A Writing Camp Q&A session with Katie Parrott (staff writer and AI editorial lead at Every) and Kate Lee (Every's editor-in-chief), following Parrott's session on her process for writing with AI, including extracting voice patterns, identifying anti-patterns, and building effective style guides. # Video Identity Observations: Writing Camp Q&A with Katie Parrott & Kate Lee ## Channel Self-Description (Explicit Claims) ### Organizational Identity: "Every" The video opens with explicit self-description from the host: > "I am Kate Lee. I'm the editor-in-chief of Every... Katie Parrot, our staff writer." The channel/organization is presented as **Every**, a media company with a subscription model: > "Every is the only subscription you need to stay at the edge of AI." ### Claimed Scope and Product Line Kate Lee explicitly enumerates the scope of what Every offers: - **Newsletter**: "It consists of ideas, which is our newsletter" - **Courses and camps**: "courses and camps and other kinds of content like that" - **Podcast**: "Our podcast as well" - **Apps**: "so chorus, spiral, sparkle and monologue" - **Training/consulting**: "our consulting practice as well as courses that that we offer to subscribers" ### Named Content Verticals The transcript demonstrates a multi-column publication structure: - **Working Overtime** — Katie Parrott's column ("Who writes our working overtime column") - **Vibe Checks** — "day one or day zero model reviews" - **Source Code** — produced "with our Studio team" ### Demonstrated Editorial Hierarchy The video evidences named roles: an **editor-in-chief** (Kate Lee), a **staff writer** (Katie Parrott), and a **managing editor** ("our managing editor Eleanor"). ## Intended Audience ### Writers Interested in AI-Assisted Workflows The audience is addressed as people who attended or missed a prior "Writing Camp" and submitted questions: > "We did our first writing camp... about how Katie in particular uses AI in her writing process. We got so many questions... Over 50 of them in fact." Audience members are named and quoted directly (e.g., Luke, Eloisa Wolf, Jonathan Calvert, Jillian Young, Rory, Karen Henke, Danielle Green, Isaac Parker, Alexander Petri, Prasun Patel, Morgan, Manas), indicating a **live, interactive subscriber community**. ### Assumed Audience Knowledge The speakers assume familiarity with: - AI writing tools (Claude, ChatGPT, Spiral, Grammarly) - Concepts like **projects, skills, custom instructions, memory, context windows, artifacts, Markdown files** - Writing craft vocabulary (diction, syntax, structure) ## Demonstrated Content Format and Scope ### Format: Live Expert Q&A / Follow-up Session The video is a **live-streamed, themed Q&A** structured around audience-submitted questions, grouped into themes: *style guides, feeding Claude your voice, projects vs. custom instructions, memory/context/document syncing, workflow, tools/platforms, and creativity/taste*. ### Scope: Practical AI Writing Craft Demonstrated topics include: - Building and refining AI style guides - Distinguishing AI "tells" (e.g., "not X but Y, delve") - Claude projects vs. skills vs. custom instructions - Google Docs integration and workflow habits ## Cross-Promotion and Community Signals ### Events Pipeline The host points to a public events infrastructure: > "We do have an events page every.to/events, where we do have all of our upcoming camps, courses, meetups." ### Product Cross-Promotion **Spiral** is promoted as part of the subscriber bundle: > "Spiral is our AI writing assistant... it's one of the four products we offer in our bundle." An internal Markdown editor ("Proof") is mentioned as in development: "we have a Markdown editor that we're working on internally." ### Educational Offerings and Demand-Sensing A prior course is referenced — "how to write with AI... we taught it I believe it was last year" — plus stated plans: "we are actively planning more courses for this year and beyond," with an explicit request for audience demand signals. ## Creator Claims vs. Demonstrated Patterns - **Claim**: Every positions itself as a comprehensive AI-focused subscription ("the only subscription you need to stay at the edge of AI"). - **Demonstrated**: Deep, practical, craft-focused AI writing instruction delivered by named editorial staff in an interactive, community-responsive format, consistent with the claimed scope of camps, courses, and columns. # Delivery Observation: Writing Camp Q&A with Katie Parrott & Kate Lee ## Format & Roles - **Two-speaker moderated Q&A.** Kate Lee acts as host/moderator, reading audience questions and steering themes; Katie Parrott answers most substantive questions. - Thematic organization is announced explicitly: > "We are going to go through the questions in themes. Group themed groupings. So we're not going to go through them in the order in which they were asked." ## Katie Parrott's Delivery Patterns ### Conversational restarts and self-repair A very frequent pattern: starting a word or phrase, abandoning it, and restarting — visible throughout the transcript: - > "you know how you how you have your projects set up" - > "it's it's a prompt that or a set of instructions" - > "I just I found like for what like most of the writing that I do" This appears dozens of times and is a stable feature of her delivery. ### Filler and hedges - Heavy use of **"like"** and **"you know"** — she names this herself: > "just from my talking now, you can hear that I say like and um a lot" - Repeated hedge phrases before answering: > "Yeah, I I I think that that is that is totally legit"; > "Um, I I think that it depends" ### "It depends" framing Answers often begin by conditioning the answer on context: - > "it really depends on what you're doing and how and how you're using your style guides" - > "it's it's very contextual. So, it depends on like where you are in the process" - > "it depends on you know, I'm like I haven't I don't have as much experience" ### Concrete examples and named tools She habitually grounds abstract answers in specific named artifacts: - Style guides as > "just a Google Doc. It's a or it can or it can be an MD file" - Named checks: "AI check skill," "thesis check," "Katie Voice" skill - Named writers: "Jia Tolentino, John Green, Leslie Jamison" - Named books: *Made to Stick*, *Hit Makers*, *A Swim in a Pond in the Rain*, *The War of Art*, *Bird by Bird* ### Analogies as a teaching device - > "it's going from bricklaying or like construction to sculpting or molding or like pottery" - > "It's cliché to say that AI is like a junior employee, but it really is" - > "if you use not X but Y, I will come and haunt your family" (humorous threat as a memorable custom instruction) ### Honest admission of limits She repeatedly flags uncertainty explicitly rather than guessing: - > "I don't have like a ton of data that back that up" - > "I'm not 100% [clears throat] sure as far as like what integrations are and are not possible" - > "I'm not 100% like up to speed on the technical details" - On "murder board": > "I don't remember what it came up in reference to. So, if we don't know, we can move on." ## Kate Lee's Delivery Patterns - **Bridging and summarizing** after Katie's answers: "Got it," "Okay," "That's interesting," "Great. Okay." — this occurs at nearly every transition. - **Recapping themes into a takeaway**, e.g.: > "a common theme I think through through writing camp... is always like your layer of judgment... on top of what the AI generates." - **Audience management**: acknowledging commenters by name ("David Franzen just clarifies...", "Thank you for sharing that comment, uh, David") and prompting for questions near the end. - **Promotional framing** at open and close: the Every subscription pitch at the start, and the events-page/course-demand invitation at the end. ## Interaction Habits (Both) - **Laughter and audible snorts** punctuate the session, marked repeatedly in the transcript — often after jokes or a misstatement. - **Live correction**: Katie misspeaks ("projects are portable from chat to chat"), Kate catches it ("Sorry, you mean skills are portable"), and it's repaired with laughter — a pattern of immediate self/other correction. - **Audience-sourced answers**: when neither knows, they defer to comments ("If anyone in the audience knows and wants to put it in comments, please feel free"), and a commenter later answers (YAML, Notion integration). ## Isolated Moments (Not Patterns) - Kate's "I'm having a visual of a sort of like spy show or Homeland with the crazy board" — a one-off pop-culture aside. - Katie's joke about *A Swim in a Pond in the Rain*: > "I'm only saying that in order to curry brownie points with Dan" — an isolated humorous moment, though consistent with her broader self-deprecating humor. - Kate relaying the Spiral GM's answer verbatim — the only point where she delivers substantive content rather than moderating. ## Uncertainty Notes - Speaker attribution in a few overlapping exchanges (e.g., the "Right. Right. Be extremely direct and repeat yourself" line) is ambiguous; it most likely follows Kate's voice, but the transcript does not mark it definitively. - Snort/laughter markers appear without speaker labels, so distribution between speakers cannot be verified. # Observation Document: Writing Camp Q&A with Katie Parrott & Kate Lee **Video ID:** 01a0f15a-a9dd-78fd-ab78-321318361510 **Format:** Live Q&A follow-up to a prior "Writing Camp" on writing with AI, hosted by Kate Lee (Editor-in-Chief, Every) with Katie Parrott (staff writer, *Working Overtime* column). --- ## Framing and Context - **Kate Lee** opens by noting the pair received over 50 questions at the first writing camp about how Katie uses AI in her writing process, prompting this follow-up Q&A (Kate Lee). - Questions were grouped into **themes**, not answered in original order (Kate Lee). --- ## Theme 1: Style Guides and Anti-Style Guides ### One style guide per project or one global guide? **Katie Parrott:** "It really depends" on project structure. - She maintains **separate style guides per column** (*Working Overtime*, *Vibe Checks*, *Source Code*) because each has its own purpose and "vibe." - A **standalone brand-level style guide** holds Every-wide rules and is added across all projects. ### Style guide vs. skill (question from Luke) **Katie Parrott:** - Her process "evolved organically"; format matters less than content (Google Doc or MD file). - A **skill** is a prompt/instruction set invokable across chats and projects; a **project** is a dedicated space with its own context and instructions. - She prefers **projects + style guides** for "big overarching stuff" — she calls her *Working Overtime* project an "environment" because of its accumulated context. - **Skills** are for smaller tasks: an "AI check" skill (flagging tells like "not X but Y," "delve") and a "thesis check" (is the thesis concrete and arguable?). - Qualification: "rapidly changing business landscape" is "an old one" — AI hasn't produced it in a long time. ### On AI tells **Katie Parrott:** - An AI tell is a diction or syntax choice that "has AI smell." - Two problems: (1) tells often convey no meaning — managing editor Eleanor's point that they "seem like they mean something on first pass" but don't; (2) they create **doubt in the reader's mind** about whether a person stands behind the content. - She frames it not as concealing AI use, but ensuring clarity and that the human's voice isn't "buried under a lot of AI speak." ### Traditional style guides (Elements of Style, AP) **Katie Parrott:** Every has its own dedicated style guide; she hasn't tried loading the full AP stylebook into a skill, calling that "maybe a little over ambitious." She notes tools like Grammarly (now part of the Superhuman universe) have built-in style capabilities. ### Refining an AI-generated style guide **Katie Parrott:** - She spends "a lot of time just in there editing." - Example: her guide over-indexed on the idea that *Working Overtime* "coins concepts" (e.g., it wrongly assumed she invented "vibe coding"), producing inauthentic "signature phrases" — she **deleted that line** from the guide. - She reorders **information hierarchy** so the AI hones in on what matters. - She recommends using AI to **create** the guide (it formats in a way AI reads well), but going over output "with a fine-tooth comb." **Kate Lee** observes a recurring theme: the writer's **layer of judgment** sits atop AI output. **Katie Parrott:** "Absolutely." --- ## Theme 2: Feeding Claude Your Voice and Samples ### Using other writers' work when you lack a defined style **Katie Parrott:** "Totally legit." She names admired writers — **Jia Tolentino, John Green, Leslie Jamison** — and feeds their essays to AI asking "what do you notice?" then compares across pieces. Key point: style must be **aggregated over many pieces** to find patterns; disparate inputs won't yield a cohesive guide. ### Courses/conference transcripts as input (Eloisa Wolf) **Katie Parrott:** Unsure — she has less experience translating between formats; her **intuition** is that spoken-to-written translation "might go sideways," but she flags she has no data to back this. However, using transcripts for **subject matter** of an article "works really well" and Every does it often. ### Written vs. verbal style (Jonathan Calvert) **Katie Parrott:** Spoken and written styles differ because we process information differently — speech is repetitive, circles the point; in writing, "say it once, say it well." ### Exploring new styles (Julie) **Katie Parrott:** **Vocabulary matters** — the more language you have to describe desired sentence structure or macro-level organization, the better you can prompt. She cautions AI "tends to regress toward a mean," requiring conscious engagement to diversify. - **Kate Lee** recalls Katie feeling stuck in a formulaic "anecdote then zoom out" opening. **Katie Parrott:** she added **multiple structures** to the guide (step-by-step process format, etc.), pulled from individual pieces, but "you still have to be vigilant" — that same day she had to tell AI directly, "we're doing this too much." - On AI as a junior employee cliché: "It's cliché to say that AI is like a junior employee, but it really is." --- ## Theme 3: Projects vs. Custom Instructions vs. Skills ### Custom instructions **Katie Parrott:** Instructions plugged into a specific place inside a project (Claude or ChatGPT) that yield consistent behavior across chats in that project. - Every's **AI editor project** has long, detailed custom instructions for consistency. - Her *Working Overtime* project keeps instructions light for flexibility, e.g., > "If you use not X but Y, I will come and haunt your family." ### Building skills **Katie Parrott:** She isn't "100% up to speed on the technical details" because **Claude has a skill-making skill** — she describes the desired outcome and Claude builds it. She knows skills involve YAML front matter but "couldn't for the life of me tell you what that is." ### Skills vs. custom instructions (Eloisa Wolf) **Katie Parrott:** The difference is **portability**. Skills work across any chat/project where enabled; custom instructions live inside one project. - (She momentarily misspoke, reversing the terms; corrected.) - **Audience comment (David Franzen):** YAML is metadata telling Claude the skill's name and purpose in a few sentences. - **Kate Lee** offers a framing: skills are **action-oriented**, custom instructions set **background context**. **Katie Parrott** agrees: instructions "run in the background," skills "do an actual task." ### Selecting skills (Alexander Petri) **Katie Parrott:** Very contextual — different skills for different stages: structure checks at outline stage; AI check at line-level final edits (or preemptively during generation). She stresses: **"You can't really get out of learning the mechanics and best practices of writing"** — you must be able to articulate structure, diction, and style rules. ### Should writing guides be skills rather than instructions? (Prasun Patel) **Katie Parrott:** Her style guide is **neither** — just a Google Doc — to keep the project flexible across writing, editing, and research. She does have a "Katie Voice" skill, mostly for social, whose output tends toward self-deprecating humor. Acknowledges **trial and error** is needed. --- ## Theme 4: Memory, Context, and Document Syncing ### Memory (Jillian Young) **Katie Parrott:** Memory has "come a long way" in 6–12 months; she defaults to **memory on**, finding more remembered context improves writing outcomes. Qualification: Every published a piece about when *not* to enable memory; distinct writing verticals may be a case for turning it off. ### Style guide updates **Katie Parrott:** Claude refers to the **updated Google Doc** directly and applies changes "pretty immediately." Her writing workflow: toggle between chat and Google Doc; tell Claude "there's a new intro, what do you think now?" She praises the Claude–Google Docs integration. ### When to start a new chat (Rory) **Katie Parrott:** The biggest problem is **running out of context**, which she says usually happens when she's "flailing" — hitting max context before a happy draft signals she didn't know what she wanted to say. With memory, a fresh chat can pick up prior context. - **Kate Lee** relays an audience comment: syncing requires enabling the **Google Workspace integration**. --- ## Theme 5: Workflow and Writing Process ### AI interviewing you (Laurie Leach) **Katie Parrott:** Go-to prompt: > "Ask me questions one at a time to extract what I think about this topic." She always stipulates **one question at a time** to avoid overwhelm. The prompt can stay light because the background style guide carries context. ### Editing in Claude (Karen Henke) **Katie Parrott:** Claude isn't built as a writing tool; you must copy into a Google Doc for editability, though someday it may have an editable canvas. ### Losing liked rewrites **Katie Parrott:** Scroll up the chat, or use **artifact version history** to recover earlier versions. Artifacts aren't required — usually appear for long-form content, not single paragraphs. ### "Locked lines" (Karen Henke, Danielle Green) **Katie Parrott:** The Google Doc serves as her locked, controlled space; alternatively you can tell AI to keep/change specific parts — possibly more relevant in the terminal/Claude app than chat. --- ## Theme 6: Tools and Platforms ### Spiral's built-in style handling (David Friend) **Kate Lee** (relaying the Spiral GM): Spiral's system prompts are written from **Every's editorial principles**; users add writing samples; discouraged phrasing includes em dashes, "reporter voice," "learn the secrets," marketing hype, certain opening patterns — with ongoing improvement planned. **Kate Lee** adds she deletes such tells immediately in her own editing. ### Why start in Claude rather than Spiral? (Pav) **Katie Parrott:** Honest answer — when she built her workflow, Spiral was a "plug-and-play input-output" tool, not today's writing agent. She appreciates Spiral's three-option iteration but stays with Claude due to **muscle memory and switching costs**; she intends to use Spiral more. ### Claude Code / Co-work (Isaac Parker) **Katie Parrott:** She's experimenting toward a more flexible, personalized **agent version** of her process. Main adjustment: working with **MD files and folders on her machine** — same context (guides, skills), different location. ### MD files **Katie Parrott:** Stands for **Markdown** — a format AI "really likes to work with." Every is building an internal Markdown editor ("Proof," spearheaded by Dan). She cites a Twitter observation: "who would have predicted that the most powerful tool in the AI world would be a markdown file." ### Notion (Prasun Patel) **Katie Parrott:** Doesn't know — calls an audible. **Kate Lee** invites audience input; **David Friend** (comment) confirms Notion can be connected to Claude Co-work/Code. ### Other tools (Luke) **Katie Parrott:** Claude ecosystem + Google Docs is her effective set. She notes purpose-built content-marketing tools and enterprise platforms with compliance support exist for other contexts. --- ## Theme 7: Creativity, Guardrails, and Taste ### Do guardrails restrict creativity? (Isaac Parker) **Katie Parrott:** "It really can" — referencing her stale intro format. Requires "constant vigilance." She's still working out, philosophically, the difference between a **signature style** and mere **repetition**. **Taste** is deciding "does this feel alive to me or not?" — a "very mushy" judgment requiring you to trust your gut. ### Murder board (Alexander Petri) - Neither speaker remembered the reference initially; **Kristen Hackler** (comment) explains it as a visual planning method mapping pieces with thread. - **Katie Parrott:** For her typical ~1,500-word pieces, **outlining** suffices; more elaborate planning suits more complex projects. ### Defining writing (Morgan) **Katie Parrott:** AI writing isn't "constructing sentences word by word." Her analogy: moving from **bricklaying to sculpting/pottery** — shaping raw material rather than assembling sentence by sentence. She acknowledges this is "a controversial topic" but says her output "sounds like me, that I stand behind, that I'm proud of, and it feels like writing." She invokes "clear writing is clear thinking" — she's still actively thinking, "just at a different level." ### Making AI sound human, not generic (Manas) **Katie Parrott:** "It's really just about giving it examples." Seed with admired writers if needed; replace with your own work as it accumulates. **You will always be manually involved** — roughing up language, swapping generic analogies for personal ones. She notes she spends **more time editing than writing**, and — acknowledging a self-contradiction — that with structure in place she can spend more time "dialing in the language" since she isn't laboring over the first draft. **Kate Lee** observes this "brings the editing level into the writing" in a way blank-page writing didn't, a skill she's seen Every's writers embrace. ### Future courses (David) **Kate Lee:** Every taught "How to Write with AI" last year and is **actively planning more**; demand signals will shape planning. Events listed at every.to/events. --- ## Closing Advice **Katie Parrott's final thought:** **Read** — both writing you want to produce and *writing about writing*. Learn the craft's vocabulary as a coder knows "branch" or "YAML front matter." - Example: she batched **Kurt Vonnegut's eight principles of storytelling** into a skill she can invoke for analysis — only possible because she knew the framework. ### Recommended books (Katie Parrott) - **Made to Stick** (Chip & Dan Heath) — how ideas are conveyed and stick; good for internet writing. - **Hit Makers** — content broadly; frameworks like "make the new familiar and the familiar new." - **A Swim in a Pond in the Rain** (George Saunders) — about short stories but strong on **selection of detail**; she jokes she names it "to curry brownie points with Dan." - Shout-outs: **The War of Art** (Pressfield), **Bird by Bird** (Lamott). - She encourages reading about **ideas conceptually**, not just writing mechanics, "particularly when the AI is taking care of things like whether to use an em dash or not." **Kate Lee's summary:** "Write better, read more." **Katie Parrott:** "Read voraciously." --- ## Unresolved / Open Items - Spoken-transcript-to-written-style-guide translation: flagged as intuition only, no data (Katie Parrott). - Notion–Claude integration: not answered by speakers; resolved only via audience comment. - "Murder board": speakers did not recall its camp context; definition supplied by a commenter. - Katie's stated tension between "spending more time editing than writing" and still spending more time on particular sentences — she explicitly notes she is "kind of contradicting what I just said," and the tension is preserved here as expressed. # Speaker Attribution Record: Writing Camp Q&A with Katie Parrott & Kate Lee ## Video Identifiers - **Video ID:** `01a0f15a-a9dd-78fd-ab78-321318361510` - **Title:** Writing Camp Q&A with Katie Parrott & Kate Lee - **Format:** Live-streamed Q&A (live audience questions via comments), follow-up to a prior "Writing Camp" event ## Speakers and Roles ### Kate Lee — Host / Moderator - Self-identified as **Editor-in-Chief of Every** - Role in video: introduces session, reads audience questions, moderates themes, provides promotional segments, delivers closing remarks ### Katie Parrott (also spoken as "Katie Parrot") — Featured Expert / Staff Writer - Self-identified via Kate Lee's introduction: **staff writer at Every**, author of the **"Working Overtime" column**, also writes **"vibe checks"** (model reviews) and **"source code"** with the Studio team - Role in video: primary respondent to writing-with-AI questions ### Named Third Parties Referenced (not present) - **Eleanor** — Every's **managing editor**, quoted by Katie Parrott regarding AI tells: "they seem like they mean something on first pass and then you think a little bit about them and you're like that doesn't really mean anything" - **Dan** — referenced as leading Every's internal Markdown editor ("Proof") and as recipient of Katie's "brownie points"; no direct statements - **Spiral GM (unnamed)** — quoted indirectly by Kate Lee regarding Spiral system prompts (see Provenance) - Audience commenters acknowledged by name (see below) ## Audience / Commenters Questions and comments attributed to live commenters, read aloud by Kate Lee: - Luke, Eloisa Wolf, Jonathan Calvert, Julie, Alexander Petri, Prasun Patel ("Patel"), Jillian Young, Rory, Laurie Leach, Karen Henke, Danielle Green, David Franzen, David Friend, Pav, Isaac Parker, Morgan, Kristen Hackler, Manas ## Promotional Segments (flagged) - **Opening subscription pitch** by Kate Lee: "Every is the only subscription you need to stay at the edge of AI" — describing Every's newsletters ("ideas"), courses/camps, podcast, apps (**Chorus, Spiral, Sparkle, Monologue**), and consulting/training - **Spiral product segment:** Kate Lee answers a question about Every's AI writing assistant **Spiral**, relaying the Spiral GM's statement that its system prompts are based on Every's editorial principles and discourage phrasing like em dashes, "reporter voice," "learn the secrets," marketing hype, and certain opening patterns - **Internal product mention:** an internal Markdown editor ("Proof") being developed at Every - **Course/events promotion:** reference to a previously taught course "How to Write with AI," plans for more courses, and events page **every.to/events** ## Referenced Sources and Third-Party Claims - **Tools/platforms named:** Claude (projects, skills, custom instructions, memory, artifacts, Claude Code, Claude Co-work), ChatGPT, Google Docs / Google Workspace integration, Notion (audience-confirmed connector with Claude), Grammarly (noted as "now part of the superhuman universe"), Spiral, the terminal, MD/Markdown files, YAML front matter - **Style references:** *Elements of Style*, AP Style Guide (discussed, not adopted per speaker) - **Books recommended by Katie Parrott:** *Made to Stick* (Chip and Dan Heath), *Hit Makers* (author unnamed in transcript), *A Swim in a Pond in the Rain* (George Saunders), *The War of Art* (Steven Pressfield), *Bird by Bird* (Anne Lamott); Kurt Vonnegut's eight principles of storytelling (batched into a skill) - **Writers Katie admires:** Jia Tolentino, John Green, Leslie Jamison - **Every publications referenced:** Working Overtime, Vibe Checks, Source Code columns; a recently published Every piece about when to disable chat memory - **Uncertainty noted:** Katie Parrott explicitly declined to answer the Notion–Claude integration question ("I'm not 100% sure"); audience commenter (David Friend) provided the confirmation instead ## Availability Limits - No curated metadata supplied with this source; all provenance facts derive from the transcript - "Murder board" origin was unresolvable by speakers in-session; answered by commenter Kristen Hackler after the fact - No timestamps available in the supplied transcript We are live. Hi everyone. Thank you for joining us here today. I am Kate Lee. I'm the editor-in-chief of Every. I'm here with Katie Parrot, our staff writer. Who writes our working overtime column in addition to a lot of other stuff here at Every. And we're here to do a Q&A about writing with AI. We did our first writing camp. On this past Friday about how Katie in particular uses AI in her writing process. We got so many questions. We did our best to answer them in the Q in the camp, but we got so many that we wanted to do a follow-up. To answer those questions. Over 50 of them in fact. As well as field any new ones from the stream. Just as a reminder, Every is the only subscription you need to stay at the edge of AI. It is consists of ideas. Which is our newsletter, courses and camps and other kinds of content like that. Our podcast as well. Apps so chorus, spiral, sparkle and monologue. And training so our consulting practice as well as courses that that we offer to subscribers. So Katie, I don't know if you want to do any more introduction yourself. Otherwise, we'll get we'll get into it. I think you covered the bases. So why don't we get started? Okay. So we are going to go through the questions in themes. Group themed groupings. So we're not going to go through them in the order in which they were asked. We did put them together into a couple of themes. The first theme is style guides and anti-style guides. We had a lot of questions about that. So our first question. From several people was do you need to create a style guide for each project or one style guide for all projects? Yeah. So really it really depends on what you're doing and how and how you're using your style guides. So if you have like like if you have multiple projects that you're working on like like for example me at Every. As you mentioned Kate, I write across working overtime our vibe checks which are day one or day zero model reviews and then also source code with our with our um Studio team and each of those columns is very specific in terms of its purpose. What it covers and it has its own vibe and sensibility. And so in those cases yes, I do have separate style guides for each of those kinds of publications. But there's also we also have standard rules that we want to apply across all of Every's columns, right? So I we also have sort of those brand level. Details in a standalone style guide, but that standalone style guide that it has all of Every's. Values and best practices that apply across the the columns. That's in its own style guide that gets added across all the different projects. So it really depends on you know how you how you have your projects set up and what you're using them for. Great. Okay. So Luke asks, what's the difference between a style guide and a skill for writing? Yeah. \[snorts\] So style guides is funny. All of this is sort of evolved organically for me and as new capabilities have emerged, you kind of have to fold them into your process or or reconfigure your process all together. So I started with style guides which is just again, it's just a Google Doc. It's a or it can or it can be an MD file. The actual format of the file is less important than what it contains. But my like I started with the style guide attached to a project because that was sort of the form factor that was available at the time that I started. Really writing with AI in a in a dedicated way the way that I do now. And then skills sort of came along more recently. And for those who aren't aware, a skill is a sort of a set of like it's it's a prompt that or a set of instructions that you can apply across chats. Or and across different projects. So anywhere like if you're using Claude for example, any you know any like any project any chat, you can invoke a skill. Whereas a project is a dedicated place you go that has its own its own context, its own instructions. And so the way that I think about them. You know I really prefer projects and style guides for big overarching stuff like that like I almost don't call my working overtime project a project anymore. I call it my environment because it has so much context. And it's set up for everything that I want working overtime to be. And then skills I use more for um Specific tasks with within that. So like for example, I have an AI check skill that runs a check to make sure that the content doesn't have any of those pesky those pesky signs of AI writing like not X but Y, delve and You know in an increasing or rapidly changing business landscape. Although that one that's an old one man. That's it hasn't AI hasn't done that one in a long time for me. So that's a skill that I have is like check this content and make sure it doesn't have any of those those signs. Or I have a thesis check that is like is this is this thesis concrete and arguable. So I use skills for like smaller units of work. And then projects with style guides set up. I use for um For the big picture stuff that I want to be applied consistently across whatever I'm doing. That's interesting. Yeah, I think that that excuse me that distinction between. Projects and skills is one is a question we get a lot and and how do you determine when to use one or the other. We did have a number of questions about AI tells in particular and if you can say anything more about. What you mean by tells and if you feel like there any really good skill AI tell skills that are out there whether we built them or you use anyone else's. Yeah. So an AI tell is a sign it is either a. Word like diction is the technical term from writing, but it's either a word choice or a sentence structure or syntax that we can just tell comes from AI. It just kind of has AI smell. And it's something that you know we see over and over and over again. And the reason that that's a problem. Is that well first of all a lot of times these AI tells just don't convey any meaning. They seem like our our managing editor Eleanor talks about how they like they seem like they mean something on first pass and then you think a little bit about them and you're like that doesn't really mean anything. So just empty empty words that are a waste of space is one problem with AI tells. And then another is that it just I think the big problem is that it creates doubt in the mind of the reader about whether a person actually stands behind this and whether there's someone here whether like is there anyone there behind the scenes like like standing behind this content. And so you just you know. It's not about concealing the fact that you're using AI, but it is about making sure that the content is clear, readable. Is isn't just empty air and that what you're saying as the human behind the AI doesn't get buried under a lot of AI speak. Got it. This will we will come back to this to this topic in a moment. But in terms of just back to style guides for a moment. Do you use things like elements of style or the AP style guide? How effective do you think either of those are? Well we have our own dedicated style guide at Every that sort of handles the majority of rules and I have that you know we have that as a standalone document that we're able to use. I haven't necessarily experimented with trying to get like the entire AP style book into into like a skill. That might be a little over ambitious for just like the the limitations of what a skill is really set up. To to accommodate. But I know there are a lot of different tools out there that that have some style guide capabilities built in like Grammarly which is now part of the superhuman universe has those kinds of things built in. So once you have a style guide created by the AI, how do you refine it? I spend a lot of time just in there editing. You know if there's something that I disagree with. You know I might just delete it and say like for example like in my my style guide for working overtime, when I started feeding it essays, it it picked up it it picked up a lot of phrases that it thought I invented. Like for a long time I had to remind working over at my working overtime project that I did not invent vibe coding. And but it really over indexed on the idea that that. Working overtime coins concepts. And so what started happening there is that when I was working with it, it kept reaching for coined concepts and like signature phrases that I was inventing that that just didn't like they weren't organic, weren't natural. And so I went into the style guide and just deleted it out. You know coined concept as a goal because that just didn't reflect what I wanted the writing to do. So making those kinds of decisions. Changing the hierarchy of information like if there's something really important, you know maybe I'm you know mention it you know at the top, mention it multiple times. We really want to make sure that the information hierarchy is clear Um, that the AI can read it and knows what to sort of hone in on, which is why I do recommend using AI to create the style guide because it will format it in a way that AI likes because it is AI, but then in terms of the specifics, you know, you just it's even if like the output is good, I recommend going over it with a fine-tooth comb and just making sure that there's nothing in there that's weird that then creates problems down the line. Like my coin concept line that was like a single line in the style guide. Right. \>\> That became it really really uh irritating over time. Okay. So, it seems like, you know, a common theme I think through through writing camp and probably this is is always like your layer of judgment and your layer of how you would actually use things, say things, etc. on top of what the AI generates. Absolutely. Okay. Great. Let's move into um a next theme, which is kind of getting Claude set up, feeding Claude your voice and samples um in order to create these guides. So, first question is can you feed Claude with a taste of other writers' pieces, let's say articles or tweets, when you yourself don't have a defined style? Yeah, I I I think that that is that is totally legit. Like like it's um it's something that I have done a little bit myself um for um there's a side project that I have, which is my app that like helps analyze your taste. But so, I like like I just thinking about writers that you admire. For me, it's people like Jia Tolentino, John Green, Leslie Jamison. These are all really high-level editorial non-fiction writers who have voices and perspectives and um just sentence or craft details that I really respect. And so, I you know, I do the same thing with the I've done the same thing with essays and articles from them as I do with my own articles, which is just give it to the AI and say what do you notice about this? And then give it to give it the next piece and say what are what are these two pieces have in common? I think when the the really important part when you're building a style guide and a voice is it needs to be aggregated over the course of many pieces of writing in order to find the patterns and the commonalities um because if you're giving a lot of disparate stuff that doesn't have that through line, then I don't think you're going to get us you're not going to get that cohesive style guide that produces content in a really um really consistent way. Got it. So, along those lines, in terms of what works as input, could courses and conference transcripts work as input to Claude, especially for YouTube scripts? Eloisa Wolf asks. Um, I I think that it depend like it depends on you know, I'm like I haven't I don't have as much experience like um translating from format to format. Um, so I I don't know, for example, if you input a uh a transcript from a spoken call and try to translate that into a style guide for writing, things might go sideways. There's something we talked about on the campus that there is a difference between your spoken style and your written style. And so, taking written or spoken content and trying to translate that into a written style might go sideways is my intuition, but I don't have like a ton of data that back that up. I will say though, and this is something that I that um going from using transcript to create an article like in terms of the subject matter is aces. That like that works really well. That's something that we do at Every a lot. \>\> \[snorts\] \>\> Okay. Um, speaking of of of of written versus spoken, how much should our written language style match our verbal language style? And how should audio transcripts influence identification of our communication style? This is from Jonathan Calvert. Oof. Um Again, I think there's a there's a certain difference in how we speak versus how we write just because we process the information differently. Um, so just from my talking now, you can hear that I say like and um a lot and those kinds of tells. Um, you can often be very repetitive when you're speaking and say like the same thing two or three times as you're sort of circling the point you want to make. And in writing, you want to like say it once, say it well, right? Is something that we say a lot in writing in the writing world. So, um Yeah, that's that's what I have to say about that. Fine. Um, so Julie asks, how do you explore new styles when you find yourself stuck in the same one? That is a really good question. Um, I think the big thing and this is where like vocabulary becomes really important and having the language to describe what you want, the style whether that's like in terms of style on a on a word-by-word and sentence level or like the structure like on a more macro level of like the organization of the details and ideas. I think the more that you have the language to say, you know, like I want to make the sentence structure, you know, more, you know, expansive and um you know, or like fragment this through different prisms. Like the more you experiment with language and describing the writing you're looking for, the more I think you're going to get better at prompting the AI to to deliver that style. Um, I will say it is a challenge because um AI does tend to uh you know, regress toward a mean. And so, it does take like very conscious engagement to sort of break up those patterns, bring in more variety, and a lot and it takes vocabulary to to know what you're react what what you're not happy with and what you want to see change when you're trying to diversify that style. Well, I remember when you went through this where you were like every one of my pieces is starting with this and it's, you know, starts with an anecdote and then zooms out and then and you felt like you were getting a little bit um you were getting a little bit stuck. Is there anything you did then to recalibrate your guide? Yeah, I I added I actually added a couple of different structures. Um, you know, I like I went in because like that that for the the friction to framework um uh framework is what it's called in my style guide. And that that did like emerge organically out of the course of many articles that I had written before I even started using the style guide. So, it's a real structure that is true to how I write, but it does get very repetitive. And so, I added different I let instead of like I pulled out of the aggregate into or drilled down from the aggregate into individual pieces that had different formats. And so, now I have an arsenal of different structures that I can pull from. So, it's not just friction to framework, I'm start with what's hard and end with the solution, but there's also like the pro like the step-by-step process format for when I'm like writing like how I how to do X and um different things like that. And so, but it does So, there's some stuff you can sort of program into it just by changing your documentation and your context, but then you still have to be vigilant. So, like I I've been writing an article today and it wants to start with an with a story because all so many of my articles start with stories. And I just had to straight-up tell it like we're doing this too much. Like we need to find a different way into the piece. And um that was just a piece of direction that I had to give because, you know, it's it's done to death. It's cliché to say that AI is like a junior employee, but it really is. And sometimes you just have to give it instructions of exactly what you want it to do. \>\> Right. Right. Be extremely direct and repeat yourself. Um, so we're going to move into the next theme um which is Claude projects versus custom instructions. We touched on this a little bit about the difference between um uh a a skill and a and a project. Um, or excuse me, a skill and a writing guide a style guide. Um, so first things first, what are custom instructions and can you provide an example? Yeah, so custom instructions are a are instructions that you can plug into a specific place inside your project both in Claude and ChatGPT. Um, it's slightly different the like you the user the layout, obviously, so you have to kind of go hunting depending on which service you're using. But what they allow you what they are is like a set of instructions that sort of communicate your preferences and how you want um chats in that the AI to respond to you inside that project. So, again, it's about So, a project can have a bunch of different chats in it, but if you have those custom instructions set, you're going to get a consistent behavior, a consistent kind of response. So, that can be things as like we have a an AI editor project set up at Every that had like the custom instructions are like you know, like really really long and detailed because we want that consistent behavior, that consistent output. We want to make sure that the editor is looking at the same stuff consistently. Um, and so that's really long. But then like in my working overtime project where I want it to be more flexible and I don't want it to be just an editor or just a writer, I just have it some things like if you use not X but Y, I will come and haunt your family. Um He will haunt you. \>\> those kinds of Do not do this. So, that's that's how you use custom instructions. Okay. Okay. Um, getting back to that a little bit about what we said before. Um, so what in terms of a skill, what is in the skill? So, the skill uh is is it hasn't like like I'll be honest with you. I'm not 100% like up to speed on the technical details because the beautiful thing about Claude is Claude has a skill making skill. So, \>\> \[laughter\] \>\> you know, if you want so you like for the AI check skill that I built, the hook check, the thesis check, all these different checks that I built, I told I told a Claude I want to build a skill that does X, Y, and Z and Claude with this skill enabled knows what the format needs to be. So, I know for example that it involves YAML, Y A M L, front matter, but I couldn't for the life of you tell you tell me life of me tell you what that is. It's kind of that's that sense of like um yeah, it's you know, you you kind of let the technical details go in in favor of just describing the outcome that you want and then trusting the AI to build that in the way that you need it in order for the skill to be effective. Okay. So, kind of getting at the next question which Eloisa will ask again, what's the difference between skills and custom instructions? Uh it's just it comes down to portability. So, when you set up custom instructions inside of a project, those custom instructions live inside that project. And so, if you want that kind of behavior that those custom instructions dictate, you need to go to that project and have your chat inside that project. A skill on the other hand, you can invoke in any chat anywhere within your instance that the skill is enabled in. Um so, you know, you can just say use the AI check skill. And so, for back to my use case having source code working over time and vibe check, that the AI check skill that checks for AI patterns can be invoked across all those projects and I don't have to re-upload it across like in each individual project. It just exists across all of them. So, projects are portable from chat to chat and project to project, whereas custom instructions Sorry, you mean skills are portable from chat Skills, yes. Oh my god, I'm sorry. \>\> \[laughter\] \>\> Yeah, and custom instructions um live within one project and dictate the behavior of that project. Got it. Um just picking up in comments, David Franzen just clarifies that YAML equal is metadata that tells Claude in a couple of short sentences what the skill does, its name, etc. So, thank you for that information. Um David, I also saw you had a question. We will we will get to that shortly. Um I also wonder if one other way to think about um custom instructions versus skills is like skills is like you is like action oriented. You want it to do something. You want it to to actively like check or do something, whereas custom instructions are often is like setting the the sort of context for for what something should be, but is I think of it as less kind of immediately action oriented. Maybe that's just my own way of thinking about it. No, I think that's a really good point. It kind of runs in the background and sort of ensures that everything is behaving the way that you want versus it's not doing an actual task. I think that's a good good split. So, Alexander Petri asks, what is the process by which someone selects writing skills for their desired outcomes? Well, the process by which one selects skills \>\> Someone selects writing skills for their desired outcomes. Yeah, um I think that's it's it's very contextual. So, it depends on like where you are in the process for example and um like different skills are going to be needed at different points in the drafting process. So, you know, at the outline stage you're really going to be looking for at things like the organization. So, maybe you have a skill that checks structure and make sure that, you know, information, you know, flows logically from section to section and that you're building a cohesive argument and that that might be a skill you build for the outline stage. You know, the AI check is mostly relevant when you're finishing up and you're doing like line level edits on a piece that is drafted. Um although you can also like have the like, you know, I I tell the AI to use the AI AI check when I'm having it produce copy so that it hopefully comes out clean so I don't have to run the AI check again. But so but that's an example of a skill that's more for the end of the writing process. And then there's a bunch of stuff that's happens in between. I will say this is where and this is something that we talked about at the camp is that you can't really get out of learning the mechanics and the best practices of writing. Um you really have to learn things like how to talk about structure, how to talk about diction, how to talk about all these different, you know, things details about the craft of writing so that you can communicate them to your AI and tell it this is what it as you're building a skill for let's say enforcing your like style rules. Like you need to be able to articulate what your style rules are and that ultimately comes down to how much you know, how many and like your own knowledge and then also context that you're able to bring in from like trusted sources. So, um another question about this person Patel asks, shouldn't writing guides be set up as a skill rather than custom instructions? \>\> \[snorts\] \>\> I actually have my style guide set up as neither. It's set up it's just a Google Doc um because, you know, I don't want again my project I want it to be flexible in terms of how I use it. So, I don't want to set custom instructions that cause it to just work as a writer or just work as an editor because a lot of times I'm having it research or I'm having it do these different things and I don't want the the um the instructions to get too um granular. But um again, it's skills are I you definitely could make your style into a skill. Um I I actually have another skill called Katie Voice that I use. It's more for social, but sometimes I'll use it in my writing and be like this just needs to usually what comes out of it is self-deprecating humor because that is the one thing that my style guide knows more than anything about me is that's how I like to you know, inject humor into my pieces. But um so, a lot of this is like there are some best practices that work best for me, but I think you also have to trial and error a little bit whether it makes sense to have a project set up or to design a skill that has your style guide in it. Got it. Um we're going to move on to the next uh theme which is memory uh and context and document syncing. So, in your experience, Jillian Young asks, how is memory used and defined and how does it influence outcomes? Memory is something that uh has come a long way in the last 6 months to a year. Like I know that Claude didn't used to have memory and now it does and you can sort of turn it on and turn it off. I know different people have different opinions about memory and whether to have it active or not active. So, for example, we just published a piece about why when you might not want to have memory enabled in your in chats, but um I default to having memory on at all times and so like Claude will like search chats for past references and um for me that just the more it knows about me and the more it can remember, the better I found the experience to be from a writing perspective. Um but again, like you might find that that like if you're getting con bleed over and like like again, if you have those really distinct verticals and like different kinds of writing that you're doing, then maybe there's a case for turning off right memory in those cases, but um Okay. So, okay, we'll moving on to to that super tactical question. Every time you update your style guide, do you remove the old one and attach a new one or does Claude refer to the updated Google Doc? Claude refers to the updated Google Doc. So, when you make changes to that Google Doc, at least so far as I've seen, it picks it up and applies it pretty immediately. Um so, my style guide, you know, and that's the that goes for writing, too. Like, you know, when you like connect a Google Doc to a chat in order to write, like not in the custom instructions, just in the flow of the usual chat, like that's that's honestly how I write is I'll have chat chat and I'll have a Google Doc. And so, we'll like the AI and I will go back and forth a little bit and I'll transfer what I like into the Google Doc and then I'll keep going and then I'll say like, all right, I made some changes. Let's say I like made a change to the intro. I'll just say, "Hey, there's a new intro. What do you think now?" And the and Claude can go look and see the that the the intro has changed. Um so, that's one of the benefits of Claude is just that AI that Google Doc integration is really smooth. And so, if you live in Google Docs the way I have for like my entire career, um it's a really it's a really awesome sort of connection there. Got it. Okay, so it syncs. It's great to know. Documents sync between Claude and Google Docs. Um let's go into workflow and writing process. So, at what point during a writing session do you start a new chat? Have you noticed longer sessions drifting from instructions as the context window fills up? Answer Rory asks this question. I That's a good question. I don't the bigger problem the biggest problem I run into is just running out of context. Like it'll just like Claude will just tell me like this is the max amount that a conversation can take. You need to start over. Um in my experience, that usually only happens when I'm like flailing and don't know what I want to say. Like if I'm hitting the the max context of a chat before I have a draft that I'm happy with, it's probably a sign that I was floundering for what I wanted to say and was like going back and forth and like making the chat a lot longer than it needed to be. Um, but that's where the memory comes into play because like when you do hit that end of a chat, you can just tell you can start a fresh chat and say, "Can we pick up where we left off in the last chat?" And Claude has the ability to like go read that previous chat and bring that context into the new one. Got it. Um, just going to add in terms of syncing, uh, just to be clear, we have this comment that if you want it if you want to sync your Claude with your Google Doc, um, you need to have enabled the Google Workspace integration. Um, thank you for sharing that comment, uh, David. Um, moving on into workflow and writing process. So, you talked, uh, in the camp about the about asking AI to interview you. Um, Laurie Leach asks, "How did you ask AI to interview you?" Um, I I just like one of my go-to prompts is just ask me questions one at a time to extract what I think about this topic, um, or something along those lines. I do always stipulate one question at a time, otherwise it will give me 20 questions and I'll get really overwhelmed. Um, so that's really all I say and the reason but the reason that I'm able to like keep it that light probably is because I have the style guide in the background that knows what like, you know, for example, what working overtime is and the kinds of questions it needs to ask in order to fill out a working overtime style, um, content piece. But yeah, it just starts as simple as ask me questions to find out what I think. That's great. Very very open-ended. Um, okay, you you you've already referred to having a, you know, a Claude window open and a Google Doc open and and toggling between them. Um, getting a little bit further into that, can you edit in Claude or do you have to copy and paste into a doc, Karen Henke asks? Um, you know, that's that's one of the areas that like I think a dedicated like the fact that Claude isn't really built for as a writing tool, um, becomes relevant and like maybe someday it will be and you'll have an editable canvas that you can just click in and make changes to, but right now, um, yeah, you you do need to, um, if you want to make changes, you got to copy it into a like a a Google Doc or a space where you have that editability because Claude doesn't have that built in to my knowledge. Got it. Um, so when Claude rewrites and you lose what you liked, is it possible to avoid that? Are you able to always recapture something you like from an earlier part of the chat? Yeah, so I mean, well, if it's in the chat, you can always just scroll up the chat and find it. Um, and then if you it's in an artifact, so like if you So, if you have it like an artifact is like a, um, a spot like is is like a space where the, um, where Claude shows you what it's producing for you. Um, and sometimes it can be more sophisticated like a report or like a mini app. Those are also artifacts, but like at a certain point when you're writing long form, it will just generally give you an artifact, um, and there's version history on those. So, as you make changes, you can go back to the previous version and if you lose something that you like, you can pull it back out of there. But do you Is it required to use an artifact? It's not required to use an artifact. Usually, I found that it like, um, it does that when it's got like a long form piece of content that it wants to show you like a full draft or, um, yeah, it doesn't like every like if it's just writing a paragraph, it probably won't deliver an artifact cuz that just doesn't make sense for the for the content. Got it. Um, next question from both Karen Henke and Danielle Green, "How do you use, quote unquote, locked lines that the AI can't change?" Is that a feature of Claude? I locked lines that the AI can't change. Um, I mean, that's where the Google Doc for me comes in as like the cuz like I control that and whatever outputs that I want to keep and control like I live in the Google Doc, um, but as far as, um, you know, I think you could probably tell the AI like keep this part and only change this part. Um, that might be more relevant in like using the terminal and the Claude app versus, um, the chat interface. Got it. Okay. Um, we are going to move on to tools and platforms. Um, so um, we did get one question. I'll just say David Friend sent a question, uh, that leads into this where, uh, you asked, uh, in the comments, "How much of these types of good writing style Excuse me, good writing style guides and and elimination of red flag AI phrases is built into Spiral?" So, Spiral is our AI writing assistant, um, for those of you don't know, it's one of the four products we offer in our bundle. Um, and so I actually did a little just did a little chat with the the GM of Spiral to just get a get the best answer that I could. Um, and and here's what he said, which is that, um, and and I'll answer this question now, um, the best answers that the Spiral system prompts are written based on Every's editorial principles. And then you you the user can add your writing samples to teach Spiral how to write more like you. Um, we do discourage phrasing in the system prompts like em dashes, reporter voice, learn the secrets, marketing hype, certain opening patterns. Um, and we're always going to be improving this. Um, and and, uh, and and, you know, uh, adding more as we go. Um, so that's hopefully, um, an answer to the question, uh, about Spiral. Um, those those AI tells are certainly all things that I look for in writing and, uh, or in AI outputs and, uh, and delete immediately. So, Spiral does have some of them built in. Um, Pav asks, "Is there a particular reason for starting the workflow in Claude rather than Spiral?" I will go take give that to you, Katie. Yeah, um, to be honest, I, um, Spiral is a tool that I don't use as much in my workflow just because at the time that I started working in this way with projects and style guides, Spiral was in its previous form. It wasn't the writing agent that it is today. It was more of a plug-and-play input-output kind of experience. And so, the thing that I really but the thing that I really appreciate about Spiral as it has evolved is that, um, it gives you iteration like different like it'll show you three options and you can pick one and sort of dial in. So, I think it's just a different, um, a different writing experience and it's one that I like intend like I like I'm like I'm going to be more actively involved in in using Spiral, but for me just, you know, the the Claude project, it's just muscle memory. You know, you get locked into a certain way of working and then you have to like switching costs and things like that become like overhead that you don't want to have. And so, it's just like, you know, my Claude project just kind of pulls me in. So. Um, so then Isaac Parker asks, "Could your setup work better in Claude Co-work?" I have actually been kind of I'm mid I've been experimenting with Claude Code and Claude Co-work in the Claude app and as well as the terminal to sort of evolve my own writing process. This is very experimental, but I am working toward an agent version that kind of, um, just is a little bit more flexible and, um, personalized to me and like how I approach the writing process. Um, and so I'm getting I'm familiarizing myself with how this kind this process changes with Claude Code or Claude Co-work as the more user-friendly version is called. The big thing for me is getting used It's just getting used to working with MD files on your machine. Like having like having folders on your desktop or somewhere in your physical machine that you can that you put all this context in cuz like a lot of the context, like the style guides, the skills, all of those things can be used in Co-work or Claude Code as well. It's just where they live is different. And so, you kind of and you have to get used to like knowing where to go to update your style guide, where to to point Claude to find the examples that you want it to work to use as like, you know, to help triangulate your style. So, it's just a matter of familiarizing yourself with the particular features and nuances of the platform you're using, um, and, um, going and working within those kinds of constraints. So, speaking of MD files, first of all, can you explain what an MD file is? What does it stand for? \>\> \[snorts\] \>\> It stands for Markdown and it's a it's a file format that AI really likes to work with. So, for example, you know, are we allowed to talk about Proof? We have, um, Sure. Yeah, we, um, we have a Markdown editor that we're working on internally that like Dan is spearheading because specifically AI works really well with Markdown. And so, like the the symbology and all of that is stuff that it can pick up on really easily. So, um, I saw something on Twitter that's just like, who would have predicted that the most powerful tool in the AI world would be a markdown file because it's a very basic just structure that has existed for ages. So, speaking of MD files, Prasun Patel asks, can you produce MD files via Notion or connect Notion to Claude itself? I am going to call an audible on that one because I'm not 100% \[clears throat\] sure as far as like what integrations are and are not possible in term on that front. Got it. If anyone in the audience knows and wants to put it in comments, please feel free and we will answer that question. Um, Luke asks, do you use any other tools? Do I use any other tools for AI for AI writing? Honestly, you know, within the Claude ecosystem, whether I'm in in the chat interface, the chat app, or Claude code, or the terminal, between that and a Google Doc, like I think those like in terms of technical tools, those are really that's really the the tool set that I've found the most effective for my writing workflow. Um, and so that's all that I've that I've really needed, but I you know, there are a lot of tools out there, a lot that have like different use cases and different audiences. Um, you know, there are some tools that are very purpose-built for content marketing. I come from the content marketing world, so I'm more familiar with those. There are certain platforms that are really geared toward enterprise, and so they have a lot of support for like compliance and things like that. So, depending on your context and where you're writing, who you're writing for, there may be different platforms that might support provide the support that you need for your particular use case. And just David friends and confirming, yes, you can connect Notion and Claude co-work in Claude code. It is actually something I have done as well. So, just look for your connector. Um, okay, we're going to go into the next tranche of questions, which is about which is a little less tactical, a little bit more high-concept, but it's about creativity guardrails and taste, which is word we haven't actually really used just yet. Um, Isaac Parker asks that with all the guardrails for setting up your style with samples, do you feel it can restrict output creativity or novel ideas? It it really can, and we talked about that a little bit with my you know, intro format getting very stale and very repetitive. Um, and so there is a certain amount of constant vigilance, to use a Harry Potter word term, that you need to make sure that the writing um is fresh. I think something that I'm still working out for myself on a philosophical level is the difference between like a signature move or a signature style and something that's just repetitive. Because, you know, what is what is you know, stock like what is a signature versus what is like a template that's really like stale. And that's where the the word taste comes in is you just have to decide for yourself like does this feel alive to me or not? And that's like a very mushy thing, but um if you if you feel like the content's getting repetitive, you've kind of got to trust your gut, and then it's a matter of thinking creatively about okay, how can I talk to AI? What instructions can I give it to push it to think more creatively? Got it. Um, I should just say if anyone else has any additional questions, please feel free to put them in the comments. Uh, next question. I know you referred to this in the writing camp. Alexander Petri asks, is there a process by which you quote-unquote murder board something? Is it better with real people or AI? Murder board? Is it I know that that came up. I don't remember what it came up in reference to. So, if we don't know, we can move on. Yeah, let's move on. Okay. If anyone remembers, let us know in the comments. Morgan asks very high-concept question, how are we defining writing in this context? Oh man, that is a really big question. I mean, it's not constructing sentences word by word the way that you know, trad writing is, if we can call it that. You know, AI-native writing it really like to me, writing what it does is it makes writing not higher level, but more like the the analogy that I always use is it's going from bricklaying or like construction to sculpting or molding or like pottery. I like I think about like having like some raw material on a pottery wheel and you kind of shape it into the like the vase that you want. To me, that is the feeling of writing with AI versus when you're writing manually, you're like assembling it sentence by sentence. And so that I think is the difference. I still consider it writing. That's a controversial topic \>\> \[laughter\] \>\> in many circles what whether working this way can be considered writing, but um in my experience, like I like I produce work this way that sounds like me, that I stand behind, that I'm proud of, and it feels like writing. You know, clear like the there's that line, clear writing is clear thinking, and this is still I'm still actively engaged in the thinking process. I'm just thinking at a different level than that sentence that word by word, sentence by sentence level. Right. Right. And sort of you're thinking it's it's like you're sort of moved higher up a little bit, like higher up the process in terms of where that thinking happens. Kristen Hackler very helpfully explained that murder board is a way of visual planning. These are when you map out all the pieces and parts and use thread to draw the lines between all of them, moving the line as you need until the flow makes sense. I don't know if this actually did come up or maybe it was just a question in the comments. I'm having a visual of a sort of like spy show or Homeland with the crazy board of all the of all the evidence moving around. But I don't know if you have anything anything more to add there, Katie. Um you know, I I think like you know, I just I found like for what like most of the writing that I do is like that 1500 word sweet spot of like, you know, it's long-form, but not super long-form. And so I found like just outlining can is and like making sure that I'm clear on the structure that way is is is is enough to get the job done, but like you can definitely go into more elaborate forms of like planning and thinking depending on the the complexity of the project you're working at working on. Okay. Um, so our last question from Friday's camp, I will share share in a moment. And of course, if you do have additional questions, we have until 2:00, so another 12 minutes if you'd like to ask Katie or me anything else. Manas asks, how do you make AI write in your human tone without it sounding generic, which is probably like the question. \>\> \[gasps and snorts\] \>\> Um, it's really just about giving it examples. And you know, as we talked about you you you may not have your own examples to start with, and then you just seed some, you know, to get it started, you seed in some examples um of writing you admire, and then I would argue I would like say that as you produce more content that sounds like you, that you're proud of, you can replace that. But um ultimately like I think like that this is where like you will always be manually involved in writing with AI. You know, it's never going to be as simple as giving an instruction and get out an output that is perfect and sounds exactly like you. You will be sort of getting into the the guts of the writing at a certain point to you know, rough it up, make it sound more human, change an analogy that you wouldn't use to an analogy that you know, feels more specific and like personal to you. Um, those kinds of finishing touches that um that sort of like that like I spend like I almost spend more I spend more time editing than I do writing in that sense, right? Like I I spend a lot of my time being like, this isn't quite right. I'm going to switch it up to this and and things like that. Um but you know, that like I have the time to I spend more time thinking in a weird way thinking about the particular sentence like I'm kind of contradicting what I just said, but I do once I have like that like basic structure laid out, I can spend more time dialing in the language to make it sound more like me because I'm not laboring over getting that first draft on the page. Right. I think it's something we've talked a lot about here how it kind of makes each each writer sort of to bring brings in some of that editing that is maybe would have been done by someone else or not at all, but but sort of makes every writer really have to you know, go in and say, yes, this is what I meant or I want to say this differently or no, and in a way that that you didn't necessarily have to do when you were just writing with from a completely blank page. Um, it almost brings the editing level in to the writing in a way that it wasn't before, which has been a really interesting skill to see um see our writers embrace. Um, we have one question in the comments from David, which is will every release any online course on writing this way with AI? It would be cool to see a deep dive into each of the stages of the writing process. Um, so we actually have taught a course before called how to write with AI. We taught it I believe it was last year. Um, and we are actively planning more courses for this year and beyond. It was very clear to us from Friday's camp, which was just an hour, that there were many, many questions, you know, why why we are why are we are here, but about really going deep into the process and really trying to understand Katie's process. So we're we're we're planning. We'll we'll let you we'll let you know. We do have an events page every every.to/events, where we do have all of our upcoming camps, courses, meetups, anything else that we're doing. So keep your eye out for that. But the best thing to for us to know is if there is demand for this kind of thing. So if you are interested in this and and would sign up and want to take it, you know, let us know and and we will factor that into our planning. So if there are any other questions, we are here. And if not, Katie, are there any final thoughts? Anything you want to share? The thing that I want to share most is read. You know, the the more you read, the more you engage with writing, the better you are going to be at articulating what you do and don't like. And not only reading the kind of writing you want to produce, but read writing about writing. Like learn the craft of writing, learn the language of writing the same way that if you're a coder, you know, you know, you know, what a branch is or you know, what a YAML front matter is. \>\> \[laughter\] \>\> There's there's a parallel world of vocabulary and concepts and frameworks in writing that the more you're able to bring those in, the stronger your writing is going to be. So like the example that I always use um is or one of the examples is like Kurt Vonnegut has these eight principles of style of of storytelling that I have always really loved. And so I was able to take those principles, batch them into a skill, and then I can invoke Vonnegut and get, you know, an an analysis of how well my writing is aligning with those best practices. But I'm only able to get there because I'm familiar with Vonnegut. I'm familiar with these eight frame this this framework. And so the more the more ravenous you are is the word I have to to learn about writing, the more success you're going to have getting AI to produce great writing for you. So on that note, we did have just a quick question, which what are the three best books to get better at nonfiction writing? You mentioned Vonnegut's eight principles. What Someone earlier there had been a question about Elements of Style. What are the things that you would refer to? Some of my canonical texts are on this front is Made to Stick by Chip and Dan Heath. That's I found that to be really good for for internet writing in particular. What is that book about? Can you just explain? \>\> Made to Stick is about how ideas sort of like are conveyed and how they travel and like how you know, how to make ideas that will stay in the mind of the person hearing them. So that's a really good one. I have right here Hit Makers: How to Succeed in an Age of Distraction, which isn't necessarily so much about writing as as it is more content more broadly, but he talks about things about like, you know, making the new familiar and the familiar new. And like he has frameworks like that. You know, we really love another one, A Swim in the Pond in the Rain by George Saunders, which is actually I'm only saying that in order to curry brownie points with Dan, but it's actually about short stories, but it has really good things to to teach about like the selection of detail and how you like the details that you choose like create the story. And I think that there's there are really good learnings to have from that. So those are some of the books that I've like I've also shout out The War of Art by Steven Pressfield and Bird by Bird by Anne Lamott. There's a lot of really great books out there. And I would encourage people not to limit themselves just to books about writing like mechanically, but writing conceptually and just like ideas and what makes them compelling because all of that is going to be relevant in terms of writing, particularly when the AI is taking care of things like whether to use an em dash or not. Fantastic. So definitely a big theme is to write better, read more. Read voraciously. Well, that wraps up our questions. So thank you everyone for coming, for submitting these questions in the writing camp, and then coming here and spending this hour with us. Um We're really grateful. Any questions, you can always shoot a note to every and we will respond. And we're really we hope to do this again and we'll see you soon. So thank you so much. Thanks. Bye.