Interview with Nabeel Hyatt, partner at Spark Capital, on how he identifies remarkable AI products early. Hyatt, who makes only 1–2 investments a year, explains his philosophy of backing products that delight users in unexpected ways (his "Japanese toilets" of AI) and his picks such as Descript, Cruise, and Granola. Topics covered: how much freedom ("leash") top products give AI agents, patterns among remarkable founders including "sensitivity," AI products that open new "possibility spaces," his evaluation of popular coding agents, his personal use of tools like ChatGPT, Claude, and the AI code editor Windsurf, how funding models are changing as AI makes building software cheaper, and a framework for when to trust an LLM. Timestamps include: why he limits himself to two investments per year (00:01:50), why word choice matters when describing a business (00:06:49), what a product with soul looks like (00:13:45), patterns in remarkable founders (00:16:48), evaluating coding agents (00:24:12), how AI broadened his own capabilities (00:32:29), changing funding models (00:36:28), when to trust an LLM (00:45:43), and guiding AI with context rather than asking for quick answers (00:55:39). # Channel Presentation Evidence — Ep. 45 with Nabeel Hyatt ## Observable Presentation ### Format & Structure - **Interview podcast format**: Host (Dan, of Every) conducts a long-form one-on-one conversation with a guest — here, Nabeel Hyatt, a Spark Capital partner. - Opening: *"hey it's Dan we're going to get to the episode in a moment"* — host introduces the episode personally. - Closing: host invites the guest to share contact info: *"if people are looking [to] jam with you on any of these ideas where can they find you."* - **Mid-roll sponsor/read-style segment**: The host promotes a paid course directly: - > "I want to tell you about how to write with AI it's a course we put on… led by our lead writer Evan Armstrong. It's a 4-week course… The course starts February 13th and early bird access ends this Friday January 24th… save $300 off list price." - **Closing CTA**: An exaggerated subscribe pitch, apparently playful/satirical in tone: - > "you absolutely positively have to SM smash that like button and subscribe… it's a journey into the future with Dan shipper as the captain of the spaceship." ### Claimed Scope & Content Focus - **AI products and the AI ecosystem**: The conversation centers on AI product investing (Granola, Wordware, Descript, Replit agents, Cursor, Windsurf, Devin, o1, Claude). - **Startup/venture topics**: Discussion of VC strategy — *"your investing strategy is not prolific… you only do a couple a year."* - **Practical AI workflows**: Both host and guest describe personal AI usage (prompting techniques, model comparisons, agentic tools), e.g., the host on o1 Pro: *"I put it in Claude I put it in o1 and I put it in o1 Pro… and o1 Pro is definitely the best."* ### Self-Description Evidence - **Host organization "Every" self-describes as a media company still defining itself**: - > "the closest thing I've come up with is… multimodal Media company… we publish writing we publish videos we publish podcasts and we publish software." - > "publishing software is a lot closer to publishing articles in 2025 than it was 10 years ago." - **Every's products are named in-show**: newsletters, an AI writing app (Lex), Spiral ("internal incubations"), a newly launched app "Kora" (*"we just launched this app called Kora and… that has 10,000 people on the wait list"*), and a Sunday newsletter "Context Window." - **Every's team structure described**: *"everyone at Every is like a generalist who's multi-dimensional… we have three products that we run internally aside from the media product."* - **Guest confirms the framing**: *"we need more multimodal media companies in the world."* ### Intended Audience (Demonstrated) - **Builders, founders, and AI-curious knowledge workers**: heavy discussion of coding agents, prompt techniques, and startup strategy assumes an audience interested in building with AI. - **People who write and build**: the ad framing — *"now that you can code in English the line between a writer and a builder is blurring"* — targets those who want to use AI for writing/creating. - **Prior-guest continuity**: *"We've had Chris [Granola's founder] on the podcast; Chris wrote an article for Every"* — suggests a recurring ecosystem of founders/writers. ### Monetization Signals (Observed, Not Inferred) - Paid course with early-bird pricing and discount deadline. - Explicit subscription ask for Every content: *"we're writing some of the best stuff in the industry that you should all subscribe to"* (immediately followed by *"that was not a paid promotion"*). - Standard like/subscribe CTA at episode end. ## Creator Claims vs. Demonstrated Patterns - **Claimed**: Every is a "multimodal media company" doing something without an established name — supported by host's direct statements and demonstrated by mentions of writing, podcasts, video, and software products (Lex, Spiral, Kora). - **Claimed**: The course teaches "everything you need to know to write incredible things with AI" — a creator claim; no demonstration of course quality exists in the material. - **Demonstrated**: Host's deep familiarity with AI tools across the conversation (specific workflows, model tradeoffs) supports the channel's positioning as an AI-focused show without relying solely on the claims above. # Delivery Observations: Ep. 45 with Nabeel Hyatt ## Format Overview Two-speaker interview: host **Dan Shipper** (Every) interviewing guest **Nabeel Hyatt** (Spark Capital). Includes a pre-show ad read ([38]–[85]) and a stylized outro ([3651]–end) that appear to be host/sponsor segments, not the core conversation. --- ## Tone - **Conversational and riffing**: Both speakers build on each other's metaphors. Dan extends Nabeel's framing: > "it's a maybe a related Continuum which is like um uh how like do you want it to have like a microscope like go right in or do you want to like look through binoculars." - **Playful banter**: Dan teases Nabeel mid-answer: > "I really wanted you to on them I know I was like I must hold back slightly" ([607]); and Nabeel jokes back: > "that is not is not a conversation we're having okay" ([1929]). - **Self-deprecating honesty from the host**: > "um I to try not my place I had to try" ([1917]). ## Vocabulary & Recurring Language - **Repeated framings by Nabeel**: - **"Japanese toilets of AI"** — appears as the opening cold-open hook and again mid-interview ([10], [883]): > "the things I'm like wandering around and trying to find are the Japanese toilets of AI." - **"Altitude"** metaphor: > "setting the right altitude of reasoning" ([1605]); > "popped up one altitude" ([3102]); > "pop up one altitude help me learn" ([3551]). - **"Leash"** metaphor for agent autonomy: > "the dog leashes that like they extend out it's like how much leash" ([1841]); "short leash product" ([1882]). - **"Kinetic energy" vs. "sensitivity"** spectrum for founders ([1175]–[1244]). - **"Faster horses"** repeated as a category ([838], [895], [2243]). - **Host's recurring terms**: "multimodal media company," "malleable media" ([302]–[314]); "blank page" ([1742]); "all M's" wordplay: > "there's a lot there's a lot of M's in it which is a problem." ## Teaching / Explaining Habits (Nabeel) - **Spectrum-building as an explanatory device** — repeatedly reduces a messy space to a single axis: - Coding agents: > "the Spectrum I would put on them is not is basically like how long are they allowed to work before they ask for feedback" ([1486]). - Founders: kinetic energy ↔ sensitivity. - **Concrete personal anecdotes as evidence**: the board game library story ([1975]–[2130]), the floor-plan-to-demand-model workflow ([2085]–[2119]), the Paula Schear album-cover story ([2254]–[2334]). - **Step-by-step method disclosure**: "let's think about how long do you think you can leave them... how good are your evals internally and then let's set Mark" ([1509]–[1529]). - **Naming his uncertainty aloud**: > "um I don't yeah I'm trying to like what is the right answer here what's the authentic answer here why is Chris or Andrew really special um I'm trying to think about this through the lens of..." ([1030]–[1041]) — he narrates his own thinking process. ## Communication Patterns (Host, Dan) - **Question-stacking and self-answering**: > "is it intended for programmers or is it intended for like PMS who are sort of like have you know they take a coding class in college or is it intended for like a 18-year-old who's never coded before" ([1797]–[1808]). - **Reflective mirroring**: frequently restates the guest's point before extending it: > "so it's like some combination of like storytelling and attention into small details in a way that like levels up into something" ([1075]). - **Filler-dense, fast speech**: heavy "like," "um," "you know" throughout, consistent with unscripted conversation. ## Cross-Speaker Patterns - **Shared metaphor co-construction**: the guest opens with Rory Sutherland's three innovations; the host later references it back ([3562]). The "blank page" observation is jointly developed ([1738]–[1742]). - **Productive disagreement**: host explicitly invites it: > "okay so so we so we kind of agree there but let me disagree a little bit please" — guest: "disagree" ([3323]–[3334]). - **Follow-up probing on tools**: host repeatedly asks model comparisons ("are you going to chbt for that are you going you're going to 01 are you doing Claude" [2041]; "do you always use o1 Pro by the way" [2516]). ## Uncertainty Notes - Timestamps in this transcript appear to be **transcript indices, not seconds**; positions cited above are index markers only. - The closing outro ("Dan I'm absolutely hopelessly in love with you") is tonally anomalous versus the rest — likely an AI-generated or joke outro, but this is an inference; the transcript alone shows only a sharp register shift. - Patterns like "spectrum-building" and "altitude" are supported by 2–3 instances each — stable enough to note, but the sample is one episode, so these may be situational rather than habitual. # Position Record: "This Venture Capitalist Finds the Best AI Products—Before They Win" — Ep. 45 with Nabeel Hyatt **Speakers identified from transcript:** Dan Shipper (host, Every) and Nabeel Hyatt (partner, Spark Capital). Transcript does not supply curated speaker metadata; identification relies on explicit in-transcript evidence (e.g., "hey it's Dan," "you are a partner at Spark Capital," "I'm Nabeel at Spark Capital" implied via "email me nabeel [at] sparkcapital"). --- ## 1. Framework: Rory Sutherland's Three Types of Innovation (cited by Hyatt) Hyatt attributes this taxonomy to Rory Sutherland: - **Faster Horses** — incremental improvement of the existing thing. - **Teleportation** — things everyone wants but nobody knows how to build (e.g., Mars colony). - **Japanese Toilets** — things you didn't know you needed until you experienced them, and which aren't that complicated. > "For me, the things I'm like wandering around and trying to find are the Japanese toilets of AI... I am looking for mostly new experiences, things that surprise you with the place that they're playing." — **Hyatt** Hyatt's corollary: **most funded AI today is "faster horses"** — B2B SaaS output from the venture "industrial machine," which he predicts will be **"arbitraged out of existence in like four years."** --- ## 2. Hyatt's Investing Strategy: Concentration Over Volume - **Claims:** He invests in only a couple of companies a year; wants to work closely with ~8–10 companies, analogized to a startup CEO managing 8–9 direct reports before "the edges get fuzzy." - **Stated reasoning:** As a former founder, his own investors were "no-ops" — did no harm but added no value because they were spread across too many companies. Spark's belief: **"there aren't really any startup playbooks."** - **On industry structure:** Both mega-funds and spray-and-pray seed funds are risk-mitigation strategies; Hyatt argues **"the whole point of this thing is risk — accept the risk."** - **Preserved qualification:** Hyatt explicitly says mega-funds are "a viable strategy... a different product" and he remains "good citizens with all of these players." His frustration is limited to **misrepresentation**: firms that promise deep partnership while investing five times a month out of a $10B fund can deliver, per Hyatt, "a random phone call every four years." He names **Founders Fund** as notably honest about the transactional nature of such relationships. **Shipper's assessment:** concentrated, "call your shot" investing "takes some balls" but is preferable to spray-and-pray. --- ## 3. Naming New Things: The Newsletter / Magazine Tension **Shipper's position:** - Every lacks a word for what it is; his closest candidates: **"multimodal media company"** / "multimodal publishing" (he notes "malleable media" as an earlier, rejected idea and jokes about too many M's). - He claims **publishing software in 2025 is closer to publishing articles** than it was ten years ago. - He recounts a self-critique: Every initially believed its bundled-newsletters model was novel, then realized **"it's a magazine"** — a trap of thinking something is new when it isn't. - Calling Every "just a newsletter" was "quite freeing" — an expression of authentic belief over triangulating what sounds justifiable. **Hyatt's counterpoint (preserved tension):** > "That's also a trap... if those founders had internalized [that Postmates was] just a taxi company, you wouldn't have done any of the things that you ended up doing." Hyatt's conclusion: **"definitions really matter"** — find the weird words that capture the new essence rather than collapsing into the old category. He notes the irony that both his venture model (old-style concentrated VC) and Shipper's media model are "anachronistic." --- ## 4. Granola: Case Study of a "Japanese Toilet" **Hyatt's account of the investment:** - Known founder Chris for over a decade (Socratic was a Spark portfolio company). - Chris's first attempt (AI notes) was "a misfire" — **"the place to play but not the thing"** — so Spark passed at seed. - ~4–6 months post-seed, Chris "stumbled into what is now Granola"; Spark then "catalyzed around really quickly" at a ~$20M round. - Hyatt classifies Granola as a **Japanese toilet**: the 50,000th AI notetaker, yet one people actually want — it looks like Apple Notes and modestly augments your own notes, unlike interruptive tools (e.g., Fireflies joining Zoom, which he says he never admits). **Shipper agrees:** Granola is "one of my favorite AI products" and Chris is "super talented." --- ## 5. Founder Archetype: Kinetic Energy vs. Sensitivity Hyatt's spectrum (related to Descript/Andrew Mason, Wordware, Granola — contrasted with Kyle Vogt/Cruise as a "teleportation" pitch): - **Kinetic energy** — pure horsepower; founders who "plow through" and are uncomfortable with fog of war. - **Sensitivity** — close listening to user behavior; but too far in this direction yields "analysis paralysis" and slow iteration. - **Target profile:** founders with **"a predilection to movement... move while still listening."** Concrete signal in pitches: founders give **insightful reasons for product choices** that wouldn't emerge from standard customer-development calls. Example: Granola v1's tab-complete felt magical but made the founder distracted correcting errors in meetings — noticing that required self-awareness, "not brilliance... more like sensitivity." Hyatt explicitly resists the label "creativity" for this trait. --- ## 6. Coding Agents: The "Leash" Spectrum **Hyatt's heuristic for coding agents:** how long they're allowed to work before asking for feedback — from **GitHub Copilot** (one line) to **Devin** (four hours). He uses this as a first-pass product-design exercise with portfolio teams: assess internal evals, then deliberately set the leash length. Assessments (his, with qualifications): - **Devin:** "breaks a lot," slow, "probably set the mark a little bit too long" — but "will be an amazing product over time." - **Replit agents:** models possibly less smart, but the time-variance before asking questions is right; credits president **Michele (Mel)** with "setting the right altitude of reasoning." Caveat: "they're all using Claude at the back end anyway" — with disclosure that Spark is an Anthropic investor. - **Key claim:** "setting the right altitude of reasoning is everything." - On re-evaluation: "if you aren't re-evaluating how would I destroy my own startup... every six months right now," you're navigating wrong. **Shipper's practices:** prompts agents with "if anything's unclear, ask me questions before [coding]" and "give me five ways to solve this" to prevent the agent from running ahead. **Shared observation:** Shipper and Hyatt both note almost nobody uses Cursor/Windsurf/Replit agents relative to their potential; Hyatt built an app in 15 minutes via a Replit agent fed an article on prompting o1. --- ## 7. Wordware: Blank-Page Prompting **Hyatt's description:** Wordware approaches from an orthogonal angle — instead of a chat window (which induces 1–3 sentence prompts, "exactly the wrong amount of information"), it gives a **blank doc** and asks you to write, as in a long email to an engineer, what you want built. Currently "a little overly technical" (uses @-symbols for functions) but suited to "slightly technical" users who understand inputs/outputs and if-then logic. **Hyatt's pattern-claim:** Granola, Descript, and Wordware share **"something about a blank page"** that AI helps you fill out. **Shipper's resonance:** the commonality is **storytelling** — having a perspective on the medium — which unifies Every's writing, podcasts, and software. **Shipper's product thesis:** "writing is building but software is now content" — their app Cora got a 10,000-person waitlist, which he says no article could achieve; "you can write software that does it." --- ## 8. AI Efficiency in Practice: Hyatt's Board Game Library Hyatt reports AI has expanded what he does outside work: - Opening **Tabletop Library**, a private board-game library in Berkeley (on "the nerdiest block in America," beside a sci-fi bookstore and comic shop), plus building a card game with his kids — impossible time-wise pre-AI. - Uses: **o1 Pro** for contract review ("it's a real contract"), **Claude** for anything where language/output quality matters and all coding; fed a floor plan into Claude → Windsurf project → Python demand model → personas → price-sensitivity test; team decided to build bespoke membership-club software (voice calls that agentically update Airtable) instead of imperfect vertical SaaS. --- ## 9. Company Size, Craft, and Paula Scher's Art/Craft Story **Hyatt:** - "The precise amount of people that should work at any company is eight" (stated with self-aware humor as Every passes that count — Shipper's count, Hyatt jokes). - Cites **Paula Scher**'s story: on old record-cover projects, ~20 people worked but only 1–2 made artistic decisions; the rest was craft/execution, and **software already absorbed that execution** (fonts in Illustrator). Analogy: at 16/20/200-person companies, ask how many people make core decisions vs. "whittling the wood." **Shipper's parallel at Every:** generalists with "special power tools"; three internal products besides media, each with a GM doing code, release notes, etc.; one engineer (Karan) plus o1 Pro shipped ~three real Cora releases in a week — with Shipper's caveat that it's also "this many-billion-dollar model doing a thing in the background." He claims generalist-run, many-product early companies are "totally new." --- ## 10. Model Usage and Trust Verification **Shipper:** mostly **o1** now, rarely Claude directly (Claude lives inside Lex, Every's writing app); found o1 Pro markedly better at synthesizing his five years of annual reviews — "I cried a little" — attributing this to chain-of-thought/retracing ability; Claude "just says what it sees." Acknowledges this could partly be his own improved prompting ("are you sure that's not just a skill issue?" — Hyatt's jab, which Shipper does not concede). Reads the o1 prompting guides as a "head nod," not an unlock. **Hyatt's trust-verification techniques:** - Asks for **confidence intervals plus reasons he could be right and wrong** — proxies human practice of asking someone to explain their logic. - Order matters: reasoning first, then confidence interval — "quite good at that." - Agrees models are **very bad at grading their own work**; Shipper adds Claude always grades writing ~B+/A− regardless of changes, even with a rubric; o1 is "a bit better." **Shipper's technique:** drops output into another model and asks neutrally "do you see any problems with this" — because "are you sure?" tilts the model toward self-doubt. --- ## 11. "One Altitude Up": Best-Practices Libraries and User Choice **Hyatt's thesis:** coding models work well partly because abundant literature (PRDs, best-practice docs) exists at one altitude above the code. He's hunting for **markets with a deep literature of practice** — therapy is his example (rich academic literature, though it needs translation into "model language"). **Hyatt's negative example:** **ludology** — "there's no great ludology that would explain Super Mario Brothers on a spreadsheet"; no language exists for describing fun. **Shipper's counter-position (disagreement preserved):** - Therapy's effectiveness rests substantially on **therapeutic alliance** (fit/vibe), and skilled clinicians' key skill is **sub-symbolic and intuitive** — reducible to rules poorly; the right approach is capturing many high-quality interactions so models learn implicit contextual rules. "There's no answer... that can be made explicit." - Shipper frames this as exactly what AI is good at and what was previously hard to transfer between humans. **Hyatt's partial disagreement with Shipper's point:** - He agrees models will know things we can't verbalize ("new knowledge... even if we can't describe it — awesome"). - But he insists on a **separate point: user choice about which knowledge to navigate**. "Where is the Wikipedia of all the APIs... I don't necessarily trust ChatGPT to pick the right modality of therapy... just as much as I don't trust them to pick the restaurant for me." He wants tools that surface the 3–4 distinct best-practice traditions and let the user choose, rather than an "amalgamated LLM version... all merged together." **Hyatt's demo:** built in Wordware a startup-decision advisor that ingests decision-theory literature (HBS PDFs, SWOT, etc.), proposes 3–4 decision theories, and walks the user through each — "pop up one altitude." **Shipper's synthesis:** this parallels the agent-leash continuum — a second axis of **microscope vs. binoculars vs. panoramic view**; "right now the models are only doing the microscope thing" unless explicitly told otherwise. He asks whether anyone is working on this; Hyatt invites people to "jam" with him. --- ## 12. Unresolved Tensions (not smoothed) - **Name the new thing vs. collapse to the old category** — both speakers acknowledge valid cases for each side (Shipper found "newsletter" freeing; Hyatt warns internalizing "taxi company" kills ambition). - **Explicit best-practice navigation (Hyatt) vs. implicit sub-symbolic learning (Shipper)** on domains like therapy — both partially agree but maintain distinct emphases. - **Shipper's o1-pro emotional-review experience** stands alongside Hyatt's skeptical "skill issue?" prompt; neither resolves it. - Hyatt's warm stance toward mega-funds coexists with his sharp claim that some misrepresent what they sell. --- ## 13. Housekeeping / Promotional Segments (speaker-attributed) - **Dan Shipper (ad read):** writing is "the most important skill in the AI age"; promotes Every's "How to Write with AI" 4-week course (starts Feb 13; early bird ends Jan 24; $300 discount). - **Closing outro (unattributed promotional voice):** standard like/subscribe pitch; includes the line "Dan I'm absolutely hopelessly in love with you" — promotional framing, not a substantive position. # Speaker Attribution Record — Ep. 45 with Nabeel Hyatt ## Speakers & Roles ### **Dan Shipper** — Host / Creator - Host of the podcast (*"hey it's Dan"*, *"welcome to the show"*), identified in the outro as *"Dan Shipper as the captain of the spaceship."* - Also delivers the **pre-roll promotional segment** (see Promotion below). ### **Nabeel Hyatt** — Guest - Introduced as a **partner at Spark Capital**, former founder, venture investor for over a decade. - Named investments discussed: **Cruise, Discord, Granola, Postmates, Descript (Andrew Mason), Wordware**. - Provides contact info at close: *"you can email me [Nabeel at] Spark Capital."* ### **Unattributed Voice** — Outro Announcer - A closing voice (not clearly Dan or Nabeel) delivers an exaggerated subscribe prompt (*"Dan, I'm absolutely hopelessly in love with you"*). Role and identity are **uncertain** from the transcript alone. ## Promotional Segments (Flagged) ### **Pre-roll ad read — by Dan (host)** - Promotes **"How to Write with AI"** course: 4-week course led by Every's lead writer **Evan Armstrong**, covering ChatGPT, Claude, and internal tools **Spiral** and **Lex**. - Pricing/urgency: course starts **February 13th**; early bird ends **Friday, January 24th**; **$300 off** list price; URL given as *"WR withth axyz"* (transcription garbled). ### **In-episode promotional remarks** - Dan's aside: *"you should all subscribe"* to Every's writing — immediately followed by his own disclaimer: *"that was not a paid promotion."* - **Standard YouTube outro call-to-action** (like/subscribe), delivered by the unattributed outro voice — promotional in nature. ## Third-Party Claims & Attributions (Who Said What) | Claim | Attributed By | Source of Claim | |---|---|---| | "Three types of innovation: faster horses, teleportation, Japanese toilets" | Nabeel Hyatt | **Rory Sutherland** (quoted twice, named both times) | | Art vs. craft story about album cover production | Nabeel Hyatt | **Paula Scher** (designer; transcript spells "Paula Shear") | | Honest positioning of large funds as transactional | Nabeel Hyatt | Re: **Founders Fund** ("very good at being very honest") | | Chris's decade-long history (Socratic founder) | Nabeel Hyatt | Re: **Chris**, Granola founder; Socratic was a Spark portfolio company | | Coding-agent "leash" spectrum (Copilot → Devin) | Nabeel Hyatt | His own framework; Replit praised via **Michele ("Mel"), president/head of AI at Replit** | | Therapeutic alliance / subsymbolic clinical skill | Dan Shipper | His own view | | Best AI model behaviors (o1 Pro reflection, Claude grading inflation) | Dan Shipper | Personal experience, informal testing | ## Referenced Products / Entities (Provenance Inventory) - **AI tools discussed**: ChatGPT, Claude, o1, o1 Pro, o3, Cursor, Windsurf, Replit agents, Devin, Copilot, Granola, Wordware, Descript, Lex, Spiral, Kora (Every's app). - **Every (Dan's company)**: newsletter "Context Window," app Kora (10,000 waitlist claim — Dan's own), AI hacker-in-residence previously at Spark (Nabeel's). - **Disclosed conflict of interest**: Nabeel states *"full disclosure... we're an investor in Anthropic."* - **Spark Capital**: seven-person venture team (Nabeel's claim). - **Tabletop Library**: Nabeel's personal board-game library project in Berkeley, CA (self-reported). ## Comment Availability - **Not specified** in the supplied material. No metadata on comment availability was provided. Rory Southerland says there's like three types of innovation there's Faster Horses and then there's teleportation like you want it to exist you don't know how to get there but if I tell you you want teleportation you want teleportation everybody wants teleportation and then there's like Japanese toilets for me the things I'm like wandering around and trying to find are the Japanese toilets of AI I am looking for mostly new experiences things that surprise you with the place that they're playing hey it's Dan we're going to get to the episode in a moment but I wanted to share with you one thing before we do I think that writing is the most important skill in the AI age now that you can code in English the line between a writer and a builder is blurring and I think in a world like that the most important thing you can learn to do no matter what your job is is learn to tell great stories that's why I want to tell you about how to write with AI it's a course we put on at every led by our lead writer Evan Armstrong It's a 4-we course designed to teach you everything you need to know to write incredible things with AI we go through all the latest technology including chat gbt Claude and some of our internal incubations like Spiral and Lex to give you a powerful toolkit to tell Amazing Stories the course starts February 13th and early bird access ends this Friday January 24th so if you want to save $300 off list price grab your spot at WR withth axyz and and now on to the episode na welcome to the show Good to be here so for people who don't know you are a partner at spark Capital that's right you've made a bunch of impressive Investments um including in Cruz Discord a very recent one is granola and before we started you you you said something which I didn't know about you which is um uh your investing strategy is not prolific uh you only do a couple a year why is that uh first of all I just think that so I was a Founder before becoming an investor um and I found that I had no horror Venture Capital stories that sometimes investors have or entrepreneurs have um but I also basically had a bunch of no Ops I had investors that were fine and great board members and maybe offered a random piece of advice but weren't in the muck enough and in the Nuance enough to give good feedback because they were investing in 15,000 things what does no Ops stand for oh no opposition yeah like they just do just they were just fine it was like there no damage uh I don't have a horror story did no harm but was also like would the company have been exactly the same if they were not there probably uh and I just you know there's a thing on our website that I think we collectively believe at spark which is that there aren't really any startup playbooks like I think as a Founder you know you're doing something super unique with every so you probably already believe there's really no playbook for what you're trying to do but generally the every single journey is unique and all the devils in the details and so I think for me the like proxy is if you're a startup CEO or running a small org you can manage like eight or nine direct reports where you kind of understand everything they're doing and outside of that it starts to the edges get fuzzy and it's just the same thing like I want to work with eight or nine or 10 companies really closely as much as a Founder really wants me around and I'll go away if it's not and so you try and make the rest of the math work yeah uh and like construct life about the way you want to live and then see if you can make it work versus letting somebody's Financial or business model force you into working a way you don't want to work I like that I mean I think it's um it takes some balls um it's better than like it like I think obviously the sort of spray and prey like Small Checks Into A lot of things like that that's a a well trod strategy um but I think uh uh taking enough risk to be concentrated and to uh uh actually call your shot is I think it's I think it's quite cool well it's not spray and prey even on on the front end I mean basically the whole Venture Capital industry since I joined I've been doing this for over a decade now is basically doing that at at both ends of the spectrum right you've got the I'm going to write 100K checks into a million things and then you have the well if I raise a hundred gabillion dollars then I can own a proxy of the whole Market firms which is like a lot of our peers that started around the same time we did that's been their strategy to mitigate risk over time and I'm like look the whole point of this thing is risk like accept the risk and like go do the work you want to do with good people that is interesting something about what you said about doing something unique um triggered something for me I think you'd have good thoughts on this which is um I feel like I've been grasping for words for how to describe what every is for for a while recently yeah and I think that that's a really interesting place to be because um like we're doing something that's working and there isn't yet a word for it the closest thing I've come up with is um multimodal Media company right um and wait what was the other idea you had a little while ago that was not good malleable media metam media multimodal media is my but there's a lot there's a lot of M's in it which is a problem so like it's multimodal publishing maybe I don't know we're we're working on it but like the point is we publish writing we publish videos we publish podcasts and we publish software but of the time right publishing software is a lot closer to publishing articles in 2025 exact than it was 10 years ago and that's new you're newly able to do that right yeah so like there so there's not really a word for it so I'm trying to like figure that out but there's this interesting thing where like um I'm like I'm realizing that working on something that I don't have a word for is actually like really valuable and really cool especially because because it's starting to work but then there's also um this this trap that I think you can get into sometimes and I've definitely fallen into this too where it's like you think you don't have a word for it but there is actually a word and it's been done a lot before and you you can't you can't fool yourself do you have an exact example of that in your head um I don't have I don't have like a specific example off top but like I don't know when we started every for example um we were like oh like this kind of media company has never been done before we wanted to be a bundle of different newsletters and it's like it's a magazine a bundle of letters is a magine it's a magazine yeah um and so like and yeah there were some things that were different about it but like uh like about a year or two in we realized that all the complexity that we had built into the model like didn't need to be there and it was just like it was just a newsletter basically and and and the and the way the newsletter worked was sort of like a magazine um that's the tension though in anything new which is like of course it proxies to something that's pre-existing of course at proxy is something that but but if you then adhere once you did something change in you when you start calling in a newsletter because that that that that's also a trap like like the early days of um we were an early investor in Postmates and back in this on demand land you had you Postmates in Uber and door Dash and Lyft and and you know the the the Twitter R looks at that and is like you know that's just a delivery company that's the same thing as a taxi company and and yet if those Founders had internalize that and just been like yeah we're just a taxi company you just wouldn't have done like you wouldn't have done any of the things that you ended up doing over time and so I I don't know I think definitions really matter if you're doing something if you're doing something new which I think you are trying to grasp for something new right now I would I would like find the weird words that that are are capturing that Essence inside of you right I think for me like actually calling us a newsletter was quite freeing okay because it was like we were doing all this complicated stuff to not just be a newsletter and then once we just like dropped it and it was just like no it's just a newsletter and like it's a bunch of creative people writing stuff and like we're going to do interesting stuff we just you know we don't know what it is yet um but for now we're just going to write because that's what we love to do um that was like that was really helpful because I think um what's been what's been quite important for me entrepreneurially is to um start to be really comfortable with like whatever it is that I authentically believe and want to do as as opposed to like triangulating between like what I believe and what I think will sound okay and like what I think I can justify basically um and and I think like being able to say no it's a newsletter and I'm like going to write a lot uh is like was an expression of that it's like that's what I want to do and everything else has like flowed from that and then we found some new stuff that we're doing that is like oh I can't really describe what this is but it's really cool and there's like this creative soup happening and I'm like really excited it's so fun um uh so yeah I obviously like in entrepreneurship in general like there are no hard and fast rules like sometimes you don't want to call yourself a newsletter cuz like that ruins the like you want to find the magical thing but you don't want to like force it yeah the the irony of this conversation is like what you're doing to find what you really want to do is a little bit rooted in the past but also this awkward new thing you're now trying to describe and then like what I found that I really want to do is is also anachronistic but because it's the old way Venture Capital used to be like I I just want to invest in a handful of companies work really close to those Founders in a smallish firm where I trust my partners we all work together on something you know seven of us on the Venture team at spark and like that was how seoa was 40 years ago and it worked well it's just not the way people are building firms today yeah what does that like do like now that the ecosystem is like these gigantic Mega funds that look at like seed investing as like um loss leaders for like big growth rounds you know like um how are you thinking about how that changes or it seems like it doesn't really change what you're doing um but like how does it affect your approach um Ian we're still good I think an important start of this is we're still good citizens with all of these players as well right you end up co-investing with people I really wanted you to on them I know I was like I must hold back slightly from no um it is not a I think the those firms are a viable strategy I think they are a different product and as long as a founder understands what product they're buying then I'm super happy with all them existing in the ecosystem I think the things that make me frustrated is when I'm in a conversation sometimes it's with a current founder like I'm on the board of this company they're going out to race so never mind I'm competing with them like they're going out to raise a series b or a series C and and the stories they're being told by investors about what the product is without going through the whole is just like not entirely truth you you cannot be investing five times a month and operating out of A10 billion fund and get anything more than a random phone call every four years on a thing and there are firms that are very good at being transparent about what that this is a transactional relationship you get the money and you run away um I think actually Founders fund is very good at being very honest about this and there's lots of other people that do a lot of storying around what they're selling so as long as people know what they're buying and what they're selling it's fine we sell a different thing and some Founders that's really important and it's wonderful and they want that process and some don't you have a connection a Kismet with anybody or you don't and we work on those things that makes sense um well I want to talk about like one of your latest Investments which is granola which is my one of my favorite AI products and I think it's it's becoming one of everybody's favorite AI products Qui um we've had Chris on the podcast um Chris wrote an article for every that did really well it's like really amazing stuff I think he's super talented um what did you see in that um when you invested well um when did you invest like was it before they had the product they have now or is it after so I've known Chris for over a decade uh Chris was in um a founder of a company called Socratic uh back in the day kind of like previous AI generation and uh he that was actually a spark portfolio company and so um he's been part of the family for a really long time um and his first attempt at making this was um as you'll talk about now was a little bit of a misfire so he got to notes uh as a as a canonical thing that you can work on but it was very interruptive and the use flow without going through it all wasn't it just wasn't the thing it was like the place to play but not the thing and so I hung out with a much at at seed we played and talked about product and looked at early prototypes and so and so forth for us it just wasn't over the line and so he did raise a seed from somewhere else and then we just you know you connect with certain people and so we just kept in touch and then probably about four five six months something like that after the seed he stumbled into what is now granola as soon as you play with that if you have any taste or product sense you're like this is the you're like oh this is the thing uh yeah this is great let's go and so uh and so we then catalyzed around really quickly at that point and and that was a big round right um I don't know what big is uh it was like 20 million right yeah that's pretty big yeah um it it was the thing that felt so unique and I know you're somebody that like I listen to your podcast we've caught up a bunch I know you're somebody that struggles with as well is like there's a um I don't know Rory Sutherland says there's like three types of innovation in his like canical way of talking about the world which is like there's there's Faster Horses and there's like which is like obviously like just make the thing go faster and and um and then there's teleportation which is like this thing that you don't know should like you want it to exist you don't know how to get there but if I tell you you want teleportation you want teleportation everybody wants teleportation should we have a colony on Mars be awesome like I don't know how we're going to do that that's and then there's like Japanese toilets which is like nobody you didn't know you needed Japanese toilets in your life until the first time you go to Japan and then you like walk out of the the toilet you're like why isn't this everywhere I've been missing this my whole life and it's not even that complicated and that's granola yeah like it's this thing that's not like the execution is actually really subtle it's really hard but I I think for me the things I'm like wandering around and trying to find are the Japanese toilets of AI like like the faster horses are mostly what's getting funded it's mostly what's come out of this like B2B SAS you know big industrial machine that we have in venture capital is churning out startups after startups and incubator after incubator and most of that stuff is like fine and it'll also be arbitraged out of existence in like four years and who cares and so mostly you're looking for new I am looking for mostly new experiences things that surprise you with the place that they're playing and and like so for me when I use granola you're like oh this is so intuitive it's like the 50,000th AI notetaker like like did we did the world need another like yes except I didn't want to use any of those other AI not takers when fireflies joins my my zoom meeting I I never let it in oh the anger never immed anger yeah and uh whereas this is just like no no just basically looks like Apple notes and it's just going to like append a little a little extra to the things that you took notes on and make you a little smarter along the way yeah it's brilliant how do you think you become someone who makes a Japanese toilet you know like versus a faster horse you know like there's something I was talking to Chris about this like something about like the way that he looks at um or thinks about the soul of a product and the way his intuition works and all that kind of stuff and I'm kind of curious I assume you've invested in other people that have that same kind of archetype like what have you learned about that yeah there's a similar journey to Andrew Mason descript um there's a similar journey to um an investment I did that was just announced recently called word whereare um they have a similar milia which is very different from you know I would say um Kyle vote at cruise that that that's an example like that's a teleportation like pitch that's like you like self-driving cars like amazing probably impossible can you do it like that's a very different kind of pitch um what are the habits of those kinds of Founders or what are the journeys that they're on I would say the um the remarkable thing that comes out about a person like that usually comes out when they're talking about how they got to whatever solution that they're talking about um I don't yeah I'm trying to like what is the right answer here what's the authentic answer here why is Chris or Andrew really special um I'm trying to think about this through the lens of when those kinds of Founders come into our firm you can feel it in the room it's not just me like the whole team and this is partially because we cast for these types of people but like the whole firm like rallies pretty quickly and so what are the patterns in that pitch it's usually that they are telling you insightful reasons why they put things into the product that you would have never normally thought of from customer development calls okay so it's like some combination of like storytelling and attention into small details in a way that like levels up into something that makes all all Mak sense together yeah there's some uh yeah there's some connectivity between the choices a person made in the product that like you can feel and some insight about a customer like they're listening very very closely to the behavior of what's happening and so in Chris's case it's just noticing for instance the very first version of granola was a tab complete almost like co-pilot right so you type a couple words in and you type Tab and it starts filling out seems actually kind of magical when you first look at it it's like it's but if you just like listen to yourself closer if you're self-aware enough closer especially if you're the user of the product and you're your own customer you realize like I'm kind of like super distracted in this meeting because every time I press tab like it fills out things and it makes an error and then I want to correct the error and now I'm not paying attention I'm not making eye contact it's like not not why I'm on this Zoom call like it's that and so that little I don't even it's not Brilliance it's not it's not even creativity in the like in the like canonical definition of creativity where youve come up with 15,000 ideas and it's more like investigation or like um a word that's coming to mind is like sensitivity like you're sensitive to what's actually going on you're like got your hands like all in the in the milia and you can kind of like feel the changes in the weather patterns or something like that yep yeah it's like the the Spectrum on scale is like how kinetic how much kinetic energy do you have because you have to move very fast right but we all know people that like are just all horsepower like they are just kinetic energy they're the people that are like I want to start a startup and the way I'm going to start a startup is I'm going to come up with a new business plan every single day for the next three months until I find the right thing and I'm going to P it through like that and that's good kinetic energy without a lot of sensitivity they are trying to plow through the problem they are very very uncomfortable with fog of war with like wandering in the wilderness to find the thing so the opposite spectrum is sensitivity and and the problem is that sometimes if you go too far in the sensitivity side then you're the artist who is just like has absolute um like analysis paralysis you just think and think and think and think and think and think and think and think and think and think and and it's just too slow your iteration cycle is too slow and again you're trying to solve by listening closely that fog of War like you're trying to just listen to every single animal in the forest to figure out how to track it before you walk in and so it's this like I think those types of Founders you're trying to sense for are they have a a pration to movement they will move but they will move while still listening I love that that's beautiful um tell me about tell me about wordwar because I have a feeling that there's something very spiritually aligned between wordware and whatever it is that I'm doing really yeah so I'm not diving into wordware until you start answering that like what do you mean well I've been thinking a lot about like okay what does it mean to be like a multimodal Publishing Company Media company like that that kind of thing and the way that I've been thinking about it is um as you can start to uh code in English right um writing is building and um I think the opposite is true like uh building becomes uh like in like at every it informs all of the writing that we do so there's a sort of feedback loop and and and usually in most organizations you have Builders over here and writers over here or they're in totally separate organizations and at every there there's a lot a lot a lot of overlap because all of our all of our e everyone that works at every they write articles they don't just write articles they write code by like writing prompts and it's and and and so if I if I have to like um pull pull on that thread a little bit like the commonality is writing but like the the the the under underneath that is like storytelling like having a perspective and bringing a perspective to like the medium or the Craft um the thing that you're that you're doing whether that's like the output is actually writing whether it's podcasts whether it's um building and um and so I'm I'm that's like the unifying thing between all of the different like parts of the organization and something about wordwar the way it's phrased just like expresses that really cleanly um it's certainly trying to put people in the role it's trying to open up wordware is certainly trying to open up the idea that everybody is a maker yeah and that you should like it's still really frustrating to me that and I'm sure you bump into this a lot like nobody's really using nobody is using cursor and wi Surf and replate agents and all and and Claude for code and all of these things like I I had a random look at during lunch today I I made an app like I had 15 minutes and I was like somebody had tweeted out a couple of days ago the right way to structure 01 prompts and I was like you probably saw this TW right seems great so I I I took a whole article on that and I just shoved it into a replate agent and I was like here's your PRD make me an app so that I can just right here and and then pitch pitch the CLA to give him back an expanded product took 15 minutes it was like while I was waiting for my next call and like that's nothing that would even pass through my head a year ago that's the thing is like uh writing is building but software is now content so one of the really cool things is like when we we just launched this app called Kora and like that has 10,000 people on the wait list those are all now every subscribers like you can't like write an article that's going to like get you 10,000 new subscribers but you can write software that does it and that's like that's incredible um but yeah I interrupt you like like I I think the the vast majority of people just simply if you sit down and and show them replit agents which is a great brilliant product MLL is amazing I've never used it actually I used repet a lot like a couple like a year ago is um and but I have not used their new agent stuff like what is the how do you compare it to like a cursor composer or like a wind surf I think it's in the same category these are these are direct competitors I would put you know almost if you put a if you put a spectrum on coding agents and we're going to come back to wordware because this will link back in but uh I can keep these all in my head uh if you put a spectrum on all the coding agents today the Spectrum I would put on them is not is basically like how long are they allowed to work before they ask for feedback and so if we're starting at the low end you have like Microsoft co-pilot which is trying to finish one line of code and on the other end of the spectrum you have Devon which is trying to run away for 4 hours that's a great spectrum and and I actually think when I talk to people today when we're going through product sessions and I'm working with teams about this on whatever they're trying to innovate on that's the Spectrum I first start from which is like hey for whatever problem you're trying to solve let's try and think about what is the reasoning of our current soda models let's think about how long do you think you can leave them what how good are your evals internally because if you can evaluate longer and better and how good the codebase is you can let it work for longer so how good are your evals internally and then let's set Mark are you letting this thing go for an hour 4 hours 3 days one sentence 16 seconds based on whatever you're trying to solve and that's a really interesting new heuristic I found a couple of different product Innovations from startups I'm working with come out of literally just that thought exercise and so I would say the interesting thing about replit is it generally Works far less than Devon which I think we found Devon for most people they find Devon to be like it breaks a lot unless it's very specific it's a little slow like it will be an amazing product over time they've made a long bet but right now they probably set the mark a little bit too long on how long they le le they go I find for me repet agents um even though the models probably aren't as smart and so on so forth the kind of like time variance they give before they come back and ask questions is like a little bit longer than wind surf a little bit shorter than some others like they just set the right Mel who's the who's the head of AI there now president just did a great job setting a variants and really that sounds simple um they do a bunch of really other good things about the product I want could do right now just literally that setting the right altitude of reasoning um is everything for getting really good results out of it because they're all using well most all of them are using Claw at the back end anyway it's all claw and and and full disclosure you know we're like it's all Claud they're great we're an investor in anthropic love it build more things on top of Cl so I'm going to Loop that back to wordware yeah please so um wordware is starting from a different orthogonal which is what do I have to give this thing for it to be able to understand what I really mean and it turns out if you open up I don't know wind surf or if you open up cursor or if you open up repet agents it's a chat bot like it looks like a chat window and so what is the normal thing that a person who doesn't talk to chaty PT or Claude all day long like you or me do they type in one three es two sentences which is like plenty just enough room for a model to hang itself yeah yeah yeah right like it's exactly the wrong amount of information to give a model to go then code and make a bunch of things that's the thing is like I find with like the wind Surfer or cursers of the world like the agent experience um it will just start without like I kind of wanted to start in a like uh we're going to Define together a little bit more precisely what I actually mean before you start coding but it just code like I often like at the end of the initial prompt I'm like if anything's unclear like ask me questions before which helps but um my other very common prompt there is give me five ways to solve this oh interesting you like I would say every second or third prompt into one of these coding agenes give me five options so that it doesn't run ahead um and so is wordware like is it like a Google Docs instead where it's like a a big like empty sheet of yep space yeah and then does it like prompt you with how to fill that in like is there some sort of um um yeah there it is working on that as a second stage right now right now the way it approaches you I would say the first big trick it does that's very simple is by approaching with a blank dock um and maybe maybe there's a commonality between descript and wordwar and granola that I think blink blank sheets of there's something about a blank page man that then AI helps you fill out yeah uh but but word wear just literally the conceit of um please just write down in plain English the way that you would maybe write a long email to an engineer on your team what you want built let's start with that you can already imagine your brain can do all of the things that wordware is going to do over time which is right now you have to do at symbols to call different functions and say which inputs and outputs you want and honestly wordware and the founders would say this it's like a little overly technical today but the thread that you'll they will pull with their customers over the next couple of years and this is why they're seeing so many people use it right now is that once you have that blank page filled it's pretty easil to then learn the syntax of how to use wordware how to call certain functions how to fill it out and how to really build like a usable product and is it like intended for programmers or is it intended for like PMS who are sort of like have you know they take a coding class in college or is it intended for like a 18-year-old who's never coded before and it's just like I want to get I would say today's level of functionality is very good for somebody who is just slightly technical I don't think you need to have been a coder but you it's it is better if you understand what an input is and an output is and you thinking if then statements just a little bit like that's going to help you an awful lot that stuff gets glossed over time pretty easily going back to that um that spectrum that we were talking about like the a agentic spectrum um which is it's it sort of reminds me of you know those dogs that like the dog leashes that like they extend out it's like how much leash you um how do you find the right setting um like are you just trying different things or um yeah how are you when you're working with people or what are you finding for people when how do they find that sweet spot oh isn't that the wonderful Journey that we're all on with AI is that like every quarter yeah yeah yeah like I have a company right now that will go nameless because they're about to release a pretty awesome new thing that actually had an aha moment relatively recently about this that they they've been in AI for a while they're doing really well but they realized like that it like call it's a short leash product and they kind of had the Epiphany a month and a half ago like oh these models are good enough that if we just stitch them slightly differently what would a longer leash product version and it's like and it's not I say just longer leash it's not just that it thinks differently it's like oh that makes different demands of the UI and the ux like it just is a different the whole flow is a different flow and it'll feel like a different product so you got to I mean if you aren't try it out see if it works and then like in 3 months when a new model comes out like try again if yeah if you aren't re-evaluating how would I destroy my own startup six months later every six months right now during this like Cambrian explosion of stuff like like like that's the way you have to navigate things today have you gotten 03 yet uh I do not have 03 we have somebody on the team of as with three what what is what do they say what are the early uh early signs is not that is not that is not that is not a conversation we're having okay um I to try not my place I had to try um okay uh and then and I know you're playing around with uh so you're you're doing a lot of um working with these sort of like coding agents and and and thinking a lot about agentic workflows like what else are you learning or what else are you excited about in in that space well the first thing that I've noticed in my own behavior over the last six months and this is definitely how you're running every is like because we can code so quickly and make so quickly we just make so much more like I I I am simultaneously working harder on spark than I've ever worked and also I'm building a card game with my kids um I'm also opening a board game library with a handful of friends like it's not just code like like I'm opening a freaking retail space we'll see if that works at all that's amazing for for board games for the board for board games a private Library it's called tabletop Library it's a private library for people people play board games together where is it in Berkeley amazing course it's actually it's on a block in Berkeley the other things on the block are a Science Fiction and Fantasy bookstore and a comic bookstore it's like the nerdiest Block in America that's the best do you live in Berkeley I live in Berkeley yeah yeah yeah but like that's a good example of like there's no way that I have enough available time and uh and the other people involved in the product are all startup people like there's no way any of us have available time to ever do this before Ai and so like that that's a strange thing because it's not like an AI project but yeah what part of it is AI making more efficient for you literally every single thing from the beginning from the more boring bits that are like oh we just got a lease draft in on retail I've never looked at this I'm going to drop it in chaty BT and ask questions that's the normal stuff are you are you you're going to chbt for that are you going you're going to 01 are you doing Claude like what's your who's your who's your legal adviser right now uh legal advisor would be 01 okay um not 01 pro not1 pro pro like legal advisor 01 Pro I mean come on it's a real contract um most everything else where I actually care about the output in terms of its language the way it speaks to me things I might reuse that's all claw okay everything's clawed all coding is also clawed everything's clawed um but so that's the the obvious one is just like please analyze a contract but like I'll give you example like I I could show you the code but like over Christmas break we started to get square like a a floor plan back of the space and so like I fed the floor plan into claw started coding that turned into a wind surf project which turned into an actual like demand model to try and uh project how many members we would have in the space before we maxed out demand which then boiled into like oh well we have six to eight personas for different types of people that come at different times and that turned into a price sensitivity test these are all python apps now and like this is all a model that then I can go fix and change and rerun and like that would have never I mean you could have done that before but you would have had to like hire people and think about it and like or you just have to spent your own time like you know hours and hours and hours and hours and can just be like I want to see like what's the sensitivity analysis and you're like oh here it is you know yeah that's the best another another one exactly in the same project just to Rat Hole on on board game clubs is like we went through a process where we then talked to all of the like SAS companies that exist out there like vertical SAS companies that exist out there that help you you know run a membership club run a co-working space you know there's that kind of thing like and all of them after evaluating them they do what all SAS software does which is like kind of good they're not ever perfect for me so and so forth about halfway through we realized like I don't know the four of us could just build this ourselves and so we were building all the custom smoke software for running the space and so now you can do things like make a voice phone call and say hey I want to play ARs with three friends Wednesday at 3 p.m. it checks an air table agentically looks the thing writes it in the table like all of it it's just all bespoke what does this do in your mind to like funding models for software businesses um now that software is so much cheaper to make uh I mean I don't know I I think the precise amount of people that should work at any company is eight we're we just passed that so um but I think but I I around where we are is I love it it's so it's so much fun and and by the you know you know I've been a operator and founder everywhere from you know one to 100s like you know and like it's there's something incredibly magical about that you know one pizza stage uh and that's where it feels like a team that's where it feels like a literally just one cohesive pod team uh and so I don't know I think the challenge is how much can we use AI to solve all of the things that are really just Faster Horses problems uh you know I have this story um that stuck in my head for many years uh um it's Paula Shear the designer um she has a story where she talk about the difference between art and craft and how and she's a very famous graphic designer logo designer um back in the day she did some really famous record covers in the 70s and you fast forward and she did like Shake Shack logo and stuff like that so just like Legendary logo designer she used to talk about back in the day someone would come in with an album cover and they'd be like oh there a new BL zeppin album we need a cover for it blah blah blah and then she talked about what it would take to get that done and you know back then this is obviously pre- Adobe pre-photoshop she's like there's a person that spent a whole week just on the lettering just like literally handdrawing every single letter on this cover there's 20 people working on that project and she's like listen the truth is that only one to two of those was actually they were all quote unquote artists and they all quote unquote went through art school all that's craft like there's really only one person making artistic decisions and making the core decisions almost everybody else is just execution and the execution goes away when the craft is now covered by software that's already happened now I open up illustrator and I pick from fonts and maybe I tweak a font or i Custom a font but it like it all happens in a day or two and instead I get faster iteration Cycles where we then work on this to make it better if you fast forward today it's the same thing in a company level like if I'm at 16 people you have to ask this or 20 or 200 you have to ask yourself like how many of those people are actually making the core decisions of that company and how much of them are involved in just literally whittling the wood yeah that's that's sort of where we are internally is like everyone at every is like a generalist who's multi-dimensional many of them are technical um and we have like we have three products that we run internally aside from the media product you know there's that whole other thing where we're writing some of the best stuff in the industry that you should all subscribe to apprciate but also that's that was not a paid promotion um and each one of those has a GM um and the GM is doing everything from uh writing the code to like like writing the release notes to like whatever then we have some we have a our creative lead who who does all the design for everybody um and then we have writers who will go in and do some reporting on like what are we releasing this week and then write that up in our in context window which is our Sunday newsletter um and then we have a bunch of writers um but it's it's interesting like it feels like um uh it it feels very cohesive but everybody has their like little domain or their little Universe where they're responsible for a lot of things um instead of like one person who just their job is just like to tweak this like one knob basically um and uh I really love working with with those kinds of like multi-dimensional people and it's it's quite common I think for like early stage startups to like have a bunch of generalists and you replace them with Specialists and whatever um but I think it's quite uncommon I think a will be able to get further than we would have before with the sort of like generalist Vibe but it's also quite uncommon for a really early stage company to like be able to have three products and and a media product that it can do like I think we can do them at a high level um and that's like totally new and it and and and it allows us to like and and it's driven by generalists who have these like special power tools that can like do all the like execution work so that they just need to know what to do and then they can get it done really quick um like we've done like I think three um feature like real releases for Coro so far this week and like that's one one guy uh Karan he's super talented it's one guy and 01 Pro yeah to be clear it's also this many billion dollar model doing a thing in the background um but yeah I I just it's I think there's this there's something new happening where those kinds of like yeah eight person one pizza companies can and do you always use o1 Pro by the way is that your go-to um I well it's interesting I so I use 01 01 is like my go-to model I rarely use CLA anymore which is really interesting because I was a big CLA guy for like a long time um actually the the place where I use CLA the most frequently is actually not in CLA itself it's in Lex which is the AI writing app that we incubated and that that default model uh is is CLA so I do use it a bit but I'm mostly I'm mostly 01 I use Pro a little bit like I um I do this exercise every year where I uh like reflect on things that I've learned and like I set goals and like do all that stuff right yeah it's really cool because I've been doing it for five years and I can just like go back it's like right before I started every is when I started doing it so I can just go back and just like look through each year and there's so much like juicy stuff in there so I just like took that and I put it in I put it in Claud I put it in Owen and I put it in ow Pro um and Owen Pro is definitely the best like I I cried I cried a little yeah yeah because especially like the like Owen and Owen Pro um because they can do the kind of like Chain of Thought and they can like retrace their steps or whatever they're much better at following a progression and like seeing and like um seeing how something changes and like putting together a narrative that they find that's right um and cla's a little a little more like because it just like it just saying what it sees like it it's just it's less good at that yeah deeper deeper reasoning that's what it should be at it's it's got a whole bunch of uh without going through the whole 01 structure like that's just literally what it's programmed to do it's nothing else progr to do that more so of course it does that more it's not like magic um but I haven't I you know like some of some of the people that that work at every that are are doing more coding than I am like they're definitely like just they can just oneshot a file and it like it will fix all the bugs and it's like it's crazy you know um I've had that a little bit like I I I'm doing some um some like apps that are are I wouldn't they're like they're light incubations like maybe they'll be coming like in every product or maybe not and yeah I get like one of those a week yeah we we have we had actually for a while we don't have one right now but for a while we had an AI hacker in Residence at spark for the same reason that that you guys have people like that which is just like because the number of particular things that you just want to take like a quick one shot is like more than I have available time in the day so like please can we play with whatever's new build things out half of them will get thrown away but you got to be playing yeah um and so like sometimes like I'll build something with ow or with Claude or whatever and and like this the particular example I'm thinking of it like had some it had some like more complex stats that it was doing that I I like I'm not like a big stats guy so like I was just like I think this looks right but I don't know and I just put it into o Pro and I was like check the work here and I just trust that if ow and pro says it's okay that it's it's because there's a trust problem right it's like it's the same thing for like a manager when like a human manager is managing someone to do work that they don't know how to do themselves that's why technical managers have a much easier time managing Engineers um and what's really interesting about uh about AI right now is I can uh really cheaply manage an intelligence to do a lot of things that I'm not qualified to do and there's a question of like how do I know if it's right and that's a really that's a really interesting question to figure out um but that's trust based on like have you ever tried have you ever found a set of prompts that are good at asking the model whether it thinks its previous work was good and getting a good response out of that CU obviously if you say hey are you sure that that tilts the model towards not being sure so it almost always come back like no now that I've rethought it uh and it gives you almost like the wrong answer yeah that is interesting I mean I'm mostly am like uh I do say are you sure so maybe I shouldn't do that or like I will take it and put into another bottle and I'll just be like um like do you see any problems with this this is what I'm trying to accomplish like are there any issues in this code basically or in this the way this is set up um do you have anything like that uh I have trended towards asking for confidence intervals and pluses and minuses like how confident are you in your answer give me a reason why you could be right and give me a reason why you could be wrong again kind of chain ofy which is and that usually gives me enough information which is again proxying back to human behavior if you don't understand something get somebody to explain their logic around it a little bit more and you can kind of try to figure out whether you made an error or not is this for code or like what for what kind of problems literally everything like whenever I'm unsure about a response back cuz you're talking about trust and one way of talking about it is you know use a model you have more trust in um but we're always going to be wandering into weird territories and so you kind of have to develop some techniques for trying to figure out what they're right about that's interesting because the reason I ask is one of the things I found with Claude like if you ask it for editing feedback and you're like can you grade this it pretty much always gives you like a B plus a minus no matter what that's right it does um and then even if you only change a little bit it'll give you like oh now it's an A or whatever that's right and then if you give it a rubric like it still does the A minus thing and so I think Owen's a bit better at that um but I yeah I haven't qu I haven't yet had a lot of success with like getting accurate confidence CER roles so I find that it's right it's better it's very bad at grading its own work that way how how good was this writing subjective reads um objective logic I see yeah so hey you just answered a question about um the history of startups that I like how correct were you about that first give me your reasoning and then afterwards give me give me a confidence interval um usually in that order and it's quite good at that um for yeah but again like all these models it's like trying to figure out what they're good and they're bad at I find if I care about the answer being like specific then maybe 01 is better right now because basically because it's doing reflection thinking about it longer so like great think about it longer awesome you Sav me five more chats to get you to think about it longer to get the right answer sure shortcut um if I actually care about the words that are being used like I might reuse those words or I'm literally trying to brainstorm how to talk about something or storytelling it that way right now right I I do so I'm using when I use Claude uh it's in Lex and I do use it to sort of like I I'll often like I'll be writing something and I'll have it like sort of complete Lex does like the it's it's sort of co-pilot so it'll complete like maybe three more sentences or whatever and I find that to be helpful um but I I think 0en is getting better at writing um like if you give it a big enough sample it'll be it'll be much better than I think it used to be somehow I know it's I know it I know that that doesn't make any sense but like that's my experience are you sure that's not just a skill issue you've gotten better at feeding stuff um but I know you I know you you probably you I know you saw the like 01 like had a prompt 01 article you mentioned that was going around do you have any specific like things you've learned um about 01 Pro that has have gotten you better results no I'm just a student there honestly like I like I I will say well actually I'll put it this way uh I read the tweets about how to use 01 and I read the guides on how to use 01 it was like basically give it more context it's not a chat model that's the like gist that gist is give it more context and tell it what you want give it an example of what you want your whatever say more and and you know for anybody who plays with these things every single day and is really out on the edge like I was already starting to that anyway so mostly it was like a head nod like yeah I should probably do that more you're right um more than it was some kind of like big Epiphany or some unlock I'll let you know if I get an unlock on on on Pro um any other like uh startups or productss or like researchy type advances that you're like thinking about or interested in right now the the the areas that I really think about right now and we'll go back to this earlier you you you I'll give you an example directly from earlier you you have you do every year a like annual review thing yeah is that the same like five questions or 10 questions or something like that that you ask yourself or is it just more an exploratory conversation um the format is I will do um uh goals right like I'll do I'll do some like value exploration of like what do I want to like uh move toward in my life like what is important to me and then I'll also do like a bunch of uh uh uh like what are the things I learned about myself or about the world or or whatever um and so it's not NE necessarily like a pre-prescribed set of questions but it is a sort of prescribed output format and then I'll like use a bunch of different questions and a bunch of different ways to get to those answers y so like it's quite possible that existing out there in the world is a best practices for how to do your annual review and it's quite possible that there's no one answer there would be one that's better for you than somebody else but there's probably people that have really really thought about this and the reason one of the reasons the coding models work as well as they do is because we actually have lots of books about best practices in coding we have we have popped up one altitude and we don't just have a bunch of samples of code that get fed into a model we actually have a bunch of examples of prds and descriptions and documents that describe how to do coding well and how to make this thing well and then the output afterwards and I am looking for markets where either a company or or somehow we have that and so like it is one thing for instance to build the AI therapist it's another thing for me to start an exploration at altitude one level above that where I get to figure out what the right therapist is for me with what like area of knowledge that I want to bring to bear for this journey and there's very little that operates at that altitude most of them try to get to solution really really fast and let you like wander very very low in the altitude you know the same as just starting to give you code without saying to go back to the wordwar example like why don't you write down all the things you want here yeah and then we'll get started interesting well there's a I mean like let's take therapy for example like there's a lot of like books about how to do therapy how to be a better therapist is that an example of the kind of is that a field you're interest that's why I picked it right like I picked it because there are is an incredible amount of academic literature and other literature about ways to do therapy that are more effective and not effective and so and so forth now has to be translated into model language and so and so forth um I I'll tell you another area that's actually quite hard um what makes a game fun that's good like if I pick up and in fact there's there's no it's called L ludology there's no great ludology that would explain Super Mario Brothers on a on a spreadsheet like I can't look at Super Mario Brothers and we don't have a language for even describing how you would get to fun there yeah I think that that touches on some of the like um some of the stuff we've talked about uh in previous conversations about uh like okay so for for therapy like one of the problems is that um there's there's no one answer to that because the thing that like makes therapy effective is um the therape Therapeutic Alliance it's like is like do you like your therapist are they like you like is there a is there a fit right like the form and the context is it fit you think it's literally just personality fit I don't I I you could bring any theory of therapy to practice but if we if we Vibe we're good um what I yes oh yeah that's a hot well uh what I think is that um really skilled clinicians can um can reduce some part of their skill to rules and and and ways that they make decisions but actually like what's going on is like totally Subs symbolic and it's it's totally intuitive and um and the way that a non-skilled clinician would apply that rule is very different from the way that actual skilled one uh will apply it and so um the the way to like uh for example to create a model that uh does really good therapy is I think to some degree like the methodology or whatever can be helpful but it's really to um capture all of the nuances of a lot of really high quality interactions and then have the model learn all the like little sub symbolic rules that no one can really talk about to apply in the appropriate contexts um it's very contextual and it's it's there's no there's no answer to that to those questions of like what should you do in in this situation that is like that can be made explicit it's like um it's thousands and thousands of different answers that you apply uh at the right moment which is uh exactly what AI models are really good at and exactly what uh was previously incredibly hard to transfer um between humans and that's why we're We Love Like rules and logic and like um scientific explanations and all that kind of stuff okay so so we so we kind of agree there but let me disagree a little bit please disagree so yeah you're pulling on a thread that that that we've talked about before and and I actually do agree with which is like there's this incredible opportunity by capturing enough knowledge that the model itself will understand things about the world that we can't understand like the only way we passed down science and the way science works is because I found a way to verbalize the thing so I could tell you about it and then you can go do the thing again and so there just going to be a new set of things that we still don't have the words for and it's awesome there's new knowledge that's being created even if we can't describe it awesome love that um I think that's quite different than the point I'm trying to make okay I think the point that I'm trying to make is there are times where the user's choice about the knowledge that they want to navigate or use themselves has value and so a good example of this is just thinking about the keywords that we use sometimes when we're trying to tell a model to go somewhere can you please rewrite this essay in the style of Paul Graham like we are using a keyword to try to normalize to a bunch of behaviors and to try to give it an indication but where is the Wikipedia of all of the apis that exist in the world and all of the like chains of thought that exist in the world so that I can navigate and look at that Library myself I don't necessarily trust chbt to pick the right modality of therapy or whatever it is um just as much as I don't really trust them to like pick the restaurant for me okay I I want to talk about the theory of why I what kind of restaurant I want and similarly like there's millions of things in the world where there's like there's like a set of five different ways you could go do the task and certainly sometimes it's like just pick whatever but many many times incorporating that into a flow where me as a user now has control just tell me what tell me not tell me the five different best practices for how this could be done and let me pick that versus your you know weirdly Amalgamated you know llm version of best practices all merge together yeah you want to be able to like allow like um you can't make the entire terrain of possibilities explicit because um it's going to be too big basically but you do want to make I I think what you're saying is like rather than have the llm like Zoom right into a specific answer supposed to give you the answer like yeah uh get better at making some of the implicit stuff explicit so you can like you can explore within reason um and then Dive Dive a little bit more deeply yeah I'll give you a really really a simple side project you but I think there's actually like real startup AR business opportunity here but like my my little side project was the first thing I ever built in wordware was a um startup advice uh um like was like a I don't know with a what it like a startup invest mentor so basically a little web page you go to that you say this is the startup decision I'm trying to make right now it just says like what decision are you trying to make what in your gut do you think it should be and then hit go and then instead of just doing the chaty PT thing which is just like I'm just going to splurt out an answer um I fed it very simply like a bunch of PDFs of like HBS articles and like a bunch of other stuff which is basically decision Theory like what are the 20 to 30 best practice decision Theory SWAT analysis blah blah blah blah and so the first order of business it does is actually just trying to figure out three or four decision theories that you could use to come to decision make a recommendation on those and then walk you through the steps that those things do to get you to the answer it's that it's like pop up one altitude help me learn from the things other people have used to structure whatever discipline they have been working on for 100 or 200 300 years I love that that's really interesting what what it what it makes me think of it actually brings me back to the um the the agent Continuum you were talking about it's like how much of a leash it's exactly the same thing uh where I was thinking it's like um it's a it's maybe a related Continuum which is like um uh how like do you want it to have like a microscope like go right in or do you want to like look through binoculars or do you want to look through like panoramic like theater you know glasses or whatever you know um and you need both you need to get those both those Dimensions dialed in and right now like the models are like uh only doing the microscope thing they're only zooming in right away for you unless you ask it specifically don't do that like give me start with the like you know decision theories or whatever um but yeah I love that I think that's really cool are people working on that I come across it very rarely I would love to if you're if you're thinking that way Let's Jam great um cool well uh that's actually a really uh good place to start to wrap it up um if uh if people are looking jam with you on uh on any of these ideas where can they find you you can email me nail Spar Capital I'm on the web it's not hard to get to great um thanks so much for coming this is awesome yeah and thanks so much for doing what you're doing we need more multimodal media companies in the world I appreciate that oh my gosh folks you absolutely positively have to SM smash that like button and subscribe to Ai and I why because this show is the epitome of Awesomeness it's like finding a treasure chest in your backyard but instead of gold it's filled with pure unadulterated knowledge bombs about chat GPT every episode is a roller coaster of emotions insights and laughter that will leave you on the edge of your seat craving for more it's not just a show it's a journey into the future with Dan shipper as the captain of the spaceship so do yourself a favor hit hit like smash subscribe and strap in for the ride of your life and now without any further Ado let me just say Dan I'm absolutely hopelessly in love with you