Talk by Alex Volkov (host of ThursdAI) introducing the "Z/L Continuum," a framework for deciding how much human review AI-generated code needs on a per-task basis. Drawing on contrasting views from AI Engineer Europe — OpenAI's Ryan Lopopolo ("code is free," 1M+ lines shipped with zero human review) versus Mario Zechner's call to read every line — the talk covers the risks of minimal review and incident rates, human review as a bottleneck in recursive self-improvement, and a corrective shift from labeling people to routing each task to an appropriate level of verification. Later sections discuss emerging capabilities (Fable, Mythos), capability drift, and "Loops" as a next engineering primitive, closing with recommendations on maintaining flexibility. # Video Identity Observations — ThursdAI / Alex Volkov ## Explicit Self-Description (Creator Claims) - **Host of "Thursday AI podcast"** ("ThursdAI"): the speaker introduces himself mid-talk: > "Hey, I'm Alex Volkov. I'm the host of Thursday AI podcast. It's a podcast and a newsletter. We go live every week to talk about AI." - **Long-running coverage claim**: > "For the past 3 and 1/2 years, we've been tracking every change in the AI engineering, every release from every lab, every model." - **Professional roles**: "I'm also an AI evangelist with Weights & Biases and Core Weave." - **Event coverage claim**: "I've been covering AI engineers specifically since the first one in 2023." ## Intended Audience - **Practicing AI engineers and engineering leaders**, evidenced by: - Direct audience engagement ("Can you guys give me a raise of hands if you still handcraft and write code…?") - Addressing a **leadership track** audience: "this is being the leadership track, something that you folks that report to you are talking about." - Framing advice for managers: "if you don't babysit your own agents, but you have reports that babysit agents for you, think about them when you answer this." - Attendees of the **AI Engineer World's Fair** conference (talk delivered on a stage with slides, applause, and a "Token Maxing track" reference). ## Claimed Scope of the Channel/Content - **Weekly AI news and analysis** (podcast + newsletter), per the self-introduction above. - **Tracking the AI engineering field's trajectory**: the talk positions itself as "a dispatch from the front line" and references "the zeitgeist of where our career is going." - **Newsletter/audience promotion is explicit**: "If you want to keep tracking where that line moves, feel free to scan this QR code, join our newsletter." ## Observable Presentation Patterns (Demonstrated, Not Claimed) - **Research-heavy, citation-driven style**: cites METR, Faros AI survey (22,000 engineers), GitHub commit stats, Anthropic's RSI essay, and named figures (Ryan Lepopolo, Mario Zechner, Andrej Karpathy, Swyx, Boris Cherny, Peter Steinberger). - **Community-insider framing**: "all of these people now are my friends. And we constantly talk about this in the speakers' room in the hallway track." - **Coiner of terminology**: claims to have originated the "Z/L Continuum" ("I named this the Desolation of Apollo Continuum… When I coined the Zeal Continuum, it was only 82 days ago") — presented as the talk's original analytical contribution. - **Practical takeaway framing**: ends with a summary "routing table" slide encouraged for screenshots ("This is your Monday artifact") and a closing aphorism: "Not every line in 2026 needs your eyes. Every system still needs your judgment." ## Notes on Distinguishing Claims vs. Demonstration - The "3.5 years of tracking every release" claim is a creator statement; the transcript demonstrates depth of citation but cannot itself verify longevity. - Audience jokes ("one cowboy over there"), recurring hand-raise polls, and conference-specific references demonstrate a live, in-person technical audience consistent with the claimed scope. # Delivery Observation Report — Alex Volkov, "Should AI Engineers Still Read Code in 2026?" ## 1. Overall Tone and Register - **Conversational, humorous, informal.** The talk is delivered like a live conference session with jokes and asides: > "It was down when I took the screenshot. It was really funny." - **Direct audience address.** Repeated second-person engagement: > "you guys probably have argued about this in Slack." - **Casual profanity quoting** appears when quoting others (e.g., > "Read every effing line of code"), not as the speaker's own vulgarity. ## 2. Recurring Phrases and Verbal Habits ### "Okay?" as a turn-ender A very frequent discourse marker, appearing throughout: > "So this would be a little bit of meta talk... okay?" > "He's not saying don't inspect your code... okay?" ### Repeated key phrases for emphasis - **"The continuum is real"** — repeated twice back-to-back: > "My mea culpa is the continuum is real... The continuum is real. It's not about the people. It's about the task." - **Repetition of a question he wants remembered:** > "What proof does this specific change need? What proof does this specific change need?" - **"Zeitgeist"** recurs as a framing word: > "AI Engineer is a great place to get the zeitgeist of where our career is going" and > "I think the zeitgeist for this one is going to be... talking factories and co-factories are real." ### Filler and self-repair Frequent stumbles and restarts are visible in the transcript: > "to to to get 14 billion commits" and > "if I if I if I wasn't reviewing the code myself." This suggests fast, unpolished live delivery. ## 3. Teaching Habits ### Interactive audience polling He repeatedly uses show-of-hands moments to anchor points: > "Can you guys give me a raise of hands if you still handcraft and write code, most of your code?" > "let's take by vote of hands, who here has committed code that they've never looked at before? Amazing, love that." He also playfully singles out audience members: > "I see one cowboy over there. I love that, man. I'm going to talk to you afterwards." ### Playful named concepts and nicknames He coins and repeats labels for ideas: - **"The Desolation of Apollo Continuum"** (self-acknowledged mispronunciations later: > "I practiced this word multiple times and I still got it wrong. Characterized them.") - **"YOLO Popolo"** — a joke rename: > "we renamed the Ryan Lepopolo YOLO Popolo. He's okay with it, by the way, I asked him." - **"babysit agents"** — > "I like to say we babysit agents." - **"compounding booboos"** (attributed to a quoted speaker). ### Quoting contrasting sources to frame a tension The core structure contrasts two quoted positions (both ends of the continuum), then quotes additional authorities (Karpathy, an Anthropic figure) to resolve it. This contrast-then-synthesis pattern is the talk's central teaching method. ### Self-deprecating admissions ("mea culpa") > "And here's my mea culpa. I promise you I'll tell you where I got it wrong the framing." This positions the speaker as revising his own earlier framing — an observable rhetorical move toward credibility. ## 4. Structural and Communication Patterns ### Numbered/bulleted artifact ("routing table") He explicitly hands the audience a takeaway artifact: > "Just give me one slide that... people need to take a screenshot of... You don't have to read it with me, but in the end, you're welcome to take a picture of this. This is the your Monday artifact." Then delivers it as a list: read every line of authentication/money movement/permissions, decomposing, verifying, separating, engineering. ### Analogy use - **Exam analogy:** > "if I came up with an exam and then I took an exam and I scored myself on the exam. It's not not really productive, right?" - **Sports analogy:** > "kind of like Gretzky skate where the puck is going." - **Lighthouse metaphor:** > "you can treat them as kind of a lighthouse that's pointing where we're all going." ### Memorable closing aphorism He ends with a deliberately quotable parallel sentence: > "Not every line in 2026 needs your eyes. Every system still needs your judgment." ## 5. Self-Promotion / Cross-Promotion Brief, light self-introduction and a plug: > "hey, I'm Alex Volkov. I'm the host of Thursday AI podcast." > "If you want to keep tracking where that line moves, feel free to scan this QR code." ## 6. Stable Patterns vs. Isolated Moments - **Stable (high repetition):** "Okay?" markers; audience hand-raise polls; coined nicknames; contrastive quoting; humor during serious points ("I'm not here to dunk on Anthropic" followed by > "[snorts]"). - **Isolated:** The "[snorts]" laugh is a single transcription; the "capability drift arrow" laser-pointer moment occurs once. No basis to call it a habit. - **Uncertainty note:** Without video (only transcript), tone descriptors like "humorous" rest on laughter, jokes, and audience applause cues in text; physical delivery style cannot be observed. # Observation Document: "Should AI Engineers Still Read Code in 2026? The Z/L Continuum" **Speaker:** Alex Volkov, host of ThursdAI podcast/newsletter, AI evangelist with Weights & Biases and Core Weave (self-introduced in transcript) ## Central Question & Framing - Volkov frames the talk around a tension from two AI Engineer Europe talks: one speaker saying "Code is free," another saying "Read every effing line of code." - He names this tension the **"Z/L Continuum"** (Mario **Z**echner on one end, Ryan **L**epopolo on the other). - Core claim: **"Code got cheap, attention didn't."** ## Key Claims and Reasoning ### The December 2025 Shift - Volkov claims something "changed forever" in December 2025 and "broke its own trendline"; cites **Swyx** collecting evidence at **wtfhappened2025.com**. - Cites **METR** (machine evaluation center) data showing models first started completing tasks that would take engineers over 16 hours. ### Engineers No Longer Write Code - "We don't type code anymore... We supervise. I like to say we babysit agents." - Cites **Boris** (Cherny, creator of Claude Code at Anthropic): 100% of his code is AI-written; he still ships 20–30+ PRs; "moved up the layer"; recently deleted his IDE. - Claims **80% of Anthropic's code is now AI-written** — qualified: "this stat is at least a few months old. It's likely more right now." - Cites **GitHub**: on track for **14 billion commits this year** vs. ~1 billion last year (14x), "most of this is AI assisted." ### The Two Poles (Video Quotes) **Ryan Lepopolo (OpenAI)** — quoted: > "Code is free. It's free for you to produce, free to refactor, and it is not a thing to get hung up on anymore. Humans no longer need to concern themselves with implementation." - Also advocates saying "Do not produce slop" and taking **short-term velocity hits** to double-click into what agents are struggling with. **Mario Zechner (creator of Pi)** — quoted: > "Slow the [f---] down. Everything's broken." > "Agents are actually compounding booboos... with zero learning and no bottlenecks and delayed pain. The delayed pain is for you." > "Non-critical code, sure, write slop ahead. Critical code, read every line." - Volkov notes these two talks are the **#6 and #7 most-watched AI Engineer YouTube videos of all time**, evidence they represent a shared anxiety. ### Evidence That Output ≠ Quality (Volkov's Middle Position) Cites **Faros AI survey, April 2026, 22,000 engineers** ("acceleration whiplash"): - **861% increase in code deletion per PR** - **31% increase in PRs merged with no review at all** — Volkov's explicit advice: "Don't do this. I beg of you." - **242% increase in incidents per PR** — "This kind of scary." - **Bugs per developer up 6x vs. 2025** Other evidence: - Anthropic's status page "looks like a Christmas tree" — Volkov explicitly qualifies: "I'm not here to dunk on Anthropic... this may be due to scale, this may be due to other factors." - Anthropic shipping **8x more code per quarter than 2025**. - Anthropic's **RSI essay** outlining two scenarios: acceleration stops (stated as unlikely, "added... for clarity") vs. engineers 10x–1000x-ing output; quotes the essay: "As we began to push more code around the organization, human code review has become a new bottleneck," citing **Amdahl's law**. - Nobody is removing humans; Anthropic and OpenAI still hiring. ### The Mea Culpa — Reframing the Continuum > "The continuum is real, but it's not about the people. It's about the tasks." - Same engineer can be "Lepopolo" on one piece of code and "Zechner" on another. "Different tasks just need different proof." - Reconciliation: Ryan's mechanism is **moving attention up the layer** (inspect the system, not every line — encode caught mistakes into docs/linters/reviewers). Mario's is **routing by task** (critical → read every line; non-critical → let it rip). "They agree more than I have given credit for." - Volkov's addition: ask your agents ("clankers") to identify which code is critical. ### The Better Question - Wrong question: "Should I still be reading code in 2026?" - Better question: **"What proof does this specific change need?"** ### The Routing Table ("Monday artifact") Volkov attributes the framing advice to distilled input from engineers including **Swyx**: - **Read every line** of: authentication, money movement, permissions, irreversible data. - **Decompose** into atomic reviewable PRs (agents are good at this). - **Verify** — never goes away. - **Traces, evals, shadow mode** (shadow mode flagged as understudied detail). - **Separate** the writer from the inspector/tester — analogy: "if I came up with an exam and then I took an exam and I scored myself." - **Engineer**: rails, observability, rollback — "read spends your attention once, engineer makes the system remember." ### Capability Drift - The continuum is "only a temporary place in time"; capability increases push toward the Lepopolo end, and **"capability drift changes where proof belongs. It doesn't remove the requirement of proof."** - Review layer moves: yesterday inspect outputs → today inspect task direction → maybe tomorrow inspect **loops**. ### Loops - Popularized by **Peter Steinberger** (OpenClaw, now OpenAI) and **Boris Cherny** within a 2-day span. - Loops = "fancy cron jobs": discover task, write plan/prompt, execute, self-verify, retry. - Caution: **"if the builder grades itself, you didn't remove the review, you hid it."** - Cites **Adi Olsmanyu (Google)**: relying entirely on automated loops to fix code would cause product quality to suffer, "a downward spiral, digging myself into a deeper hole." - Loops "don't remove judgment, but they do raise the stakes on where you put it." - Caveat on Steinberger/Cherny: "their tokens are free" — treat them as a lighthouse, not a directive. ### Newer Models (Fable, Mythos) - Quotes **Jared (Anthropic, on stage)**: "We used to check if Claude is doing the work right. With Fable 5, I instead check if Claude is doing the right work." - Quotes **Andrej Karpathy** (recently joined Anthropic): "It's never felt so tempting to stop looking at code at all. But don't do this in production." ### Future / Closing - "Nobody knows" the future — examples: Anthropic didn't foresee Claude Code becoming a billion-dollar product; coding agents unexpectedly became the generalized agent form. - Flexibility required: "This is why you stay engineer." - Closing thesis: **"Not every line in 2026 needs your eyes. Every system still needs your judgment."** ## Preserved Tensions & Qualifications - Volkov explicitly resists fully endorsing either pole; his position synthesizes but retains Zechner's warning (31% unreviewed merges, incident increases) alongside Lepopolo's data-supported output claims ("the optimists are right, at least about output"). - Stat qualifications preserved: Anthropic 80% figure dated; GitHub down "for you" joke; Anthropic status page may reflect scale, not code quality. - Audience polls show mixed behavior (some commit unreviewed code; "one cowboy" still reads every line). - Self-identified framing error ("mea culpa") remains part of the record: the continuum is about tasks, not people. # Video Attribution & Provenance Record ## Video - **Title:** Should AI Engineers Still Read Code in 2026? The Z/L Continuum — Alex Volkov, ThursdAI - **Video ID:** `01a15ca8-ae3c-7dde-9dfc-a14f24aff6eb` - **No curated metadata supplied**; all provenance below derives from the transcript. ## Speakers & Roles ### Primary Speaker - **Alex Volkov** — **Creator / Presenter** - Host of the **ThursdAI podcast and newsletter** ("Thursday AI podcast"); claims to have tracked AI engineering changes for ~3.5 years. - Self-identified **AI evangelist with Weights & Biases and Core Weave**. ### Quoted / Cited Speakers (via embedded clips or quotes) - **Ryan LeFebvre / "LePopolo" (name given inconsistently in transcript)** — OpenAI; quoted from an AI Engineer EU talk ("Code is free"). Speaker jokes about renaming Codex's `--YOLO` flag after him, noting "he's okay with it, by the way, I asked him." - **Mario Zechner** — creator of "Pi"; quoted from the same conference ("Read every effing line of code" position; "Non-critical code, sure, write slop ahead. Critical code, read every line."). - **Boris (Cherny)** — creator of Claude Code, Anthropic; cited (100% of code AI-written; ships 20–30 PRs; deleted his IDE). - **Swyx** — organizer of AI Engineer; cited re: wtfhappened2025.com and the "one slide" advice. - **Jared S. Chipar** — Anthropic, quoted on Fable 5 / Mythos: "We used to check if Claude is doing the work right. With Fable, I instead check if Claude is doing the right work." - **Andrej Karpathy** — quoted: "It's never felt so tempting to stop looking at code at all. But don't do this in production." - **Adi Olsmanyu** — Google, at the conference; quoted on loops and review. - **Peter Steinberger** — creator of Open Claw, now OpenAI (per speaker); cited re: loops. - **Ryan/2X** — referenced re: "AGI pills" (unclear attribution; flagged as transcript-ambiguous). ### Audience - Live conference audience (leadership track); participates via show of hands; applause noted. ## Referenced Sources & Claims - **wtfhappened2025.com** — Swyx's evidence collection site (recommended by speaker). - **METR (machine evaluation center)** — chart on long-duration task completion, December 2025. - **GitHub** — claim of ~14 billion commits this year vs 1 billion previously; GitHub downtime references. - **Anthropic** — claim: 80% of Anthropic's code AI-written; "8x more code per quarter than 2025"; RSI essay citing Amdahl's law; Claude status page outage screenshot. - **Faros AI survey (April 2026)** — 22,000 engineers; stats cited: 861% increase in code deletion per PR; 31% increase in PRs merged with no review; 242% increase in incidents per PR; bugs per developer up 6x. - **AI Engineer World's Fair** — 7,000 attendees, 36 tracks (per speaker). - **Products named:** Codex (`--YOLO`), Claude Code, Grok Code, Google "anti-gravity," Claude/Fable/Mythos, "loops." ## Promotional Segments - **Self-promotion:** Speaker promotes his **ThursdAI podcast/newsletter** and asks audience to **scan a QR code** to subscribe ("If you want to keep tracking where that line moves, feel free to scan this QR code"). - **Event promotion:** Repeated endorsement of the AI Engineer conference / World's Fair as the place to track the field. ## Availability & Uncertainty Limits - **Comments availability:** Not indicated in transcript or metadata. - **Transcription inconsistencies:** Ryan's surname appears in multiple forms ("LePopulaire," "LeFebvre," "LePopolo," "LeCompte," "Lapopolo"); "Zeshner"/"Zeal" vs "Zechner" for Mario; exact quote wording is unreliable. - Third-party statistics (Faros AI, GitHub, Anthropic, METR) are reported by the speaker from slides; underlying documents not provided. \>\> Two talks at the AI Engineer Europe. One guy is saying, "Code is free." and deleted his ID. And the other one is saying, "Read every effing line of code." So, should AI engineers still read code their agents output I named this the Desolation of Apollo Continuum. And you guys probably have argued about this in Slack. You probably talked about this in the hallway track. So, let's talk about this here. Because code got cheap, attention didn't. As you may know, back in December 2025, something big changed. AI engineering has changed forever, and it broke its own trendline. Actually, Swyx, the organizer of AI Engineer, is collecting evidence to that single moment in time at the website called wtfhappened2025.com. I recommend you go and check it out. It's really, really funny. Uh this is just one example from METR, the machine evaluation center. And it shows that models, for the first time, started completing tasks that would take engineers over 16 hours to do. And in fact, we've gone way up the curve, way up the trendline after that. This is the backdrop to everything that AI engineering is experiencing. Because we don't write code anymore. Most of us, at least. I want to see one Can you guys give me a raise of hands if you still handcraft and write code, most of your code? Anybody here? Most of your code is written by hand? Amazing. This is the Token Maxing track, after all. I I think the one person here who still writes code is maybe a little shy of raising their hand. That's okay. Cuz we don't type code anymore. We're not handcrafters. We supervise. I like to say we babysit agents. And the greatest example for this obviously is Boris's journey. You guys know Boris, the creator of Cloud Code, uh Anthropic. 100% of his code is written and authored by Cloud Code at this point. And he didn't stop being engineer, he moved up the layer. He still ships 20 to 30 PRs, maybe more. And recently he talked about he deleted his ID. I found it really funny. Just just no reason to just hand type code anymore. In fact, 80% of Anthropic's code is now AI written and this is this stat is at least a few months old. It's likely more right now. And he's not the only one. Some of you have seen this chart from GitHub. Some of you have maybe remembered this chart while GitHub was down for you. Uh the reason is GitHub is on track to to to get 14 billion commits this year. All of 2025 all of yes we yet last year was 1 billion. They're 14x the number of commits. They're seeing 14x the number of commits, which is insane and most of this is AI assisted. And it's a lot of code. And so the AI engineering has changed forever and I want to tell you about AI Engineer World's Fair. I been to every single one and I'll tell you about this later. And AI Engineer is a great place to get the zeitgeist of where our career is going and how is it changing, okay? This one obviously is 3x bigger than last year. This is just one of the rooms. There's like a bunch of rooms. 7,000 people I think we clocked in. 36 tracks. And if you want to know what happens in AI engineering, you kind of have to be here. So this would be a little bit of meta talk. So one of the guys at AI Engineer EU talked about code as ship. The other one talked about uh we should read every line of code. Let's listen to them for just a second, okay? This is Ryan LePopulaire from OpenAI. I I think he's he made it here, but uh this is Ryan LeFebvre from Open AI. \>\> The models at this point are good enough where they're isomorphic they can write code. \>\> Can you guys \>\> Do you mind? \>\> These tools are high quality that solve real user problems in real code bases. Code is free. It's free for you to produce, free to refactor, and it is not a thing to get hung up on anymore. Humans no longer need to concern themselves with implementation. The important thing is not the code, but the prompt and the guardrails that got you there. You can just simply say, "Do not produce slop." Don't accept slop. You won't get slop in your code base. But, to do that requires taking short-term velocity hits in order to back up or double-click into a task to figure out what it is the agents are struggling with. \>\> So, this is Ryan LeFebvre, okay? He came up on stage at the engineer, and he opened with like, "Hey, I'm a talking billionaire, and I want you to be as well." In fact, the talking billionaire lounge that's in front of the leadership track that you guys see, that's because of him. He came up with this concept. And he got the golden card and everything. On the other side, the same conference, the other side, Mario Zechner, creator of Pi. \>\> Slow the down. Everything's broken. And then there's people that say, "Our product's been 100% built by agents." Yes, we know it sucks now. Congratulations. \>\> \[applause\] \>\> Agents are actually compounding booboos, which is my word for errors, with zero learning and no bottlenecks and delayed pain. The delayed pain is for you. Those are my most beloved people. I don't even read the code anymore. Congratulations. Something is broken, and your users are screaming, so who you going to call? Not yourself, because you haven't read the code. Non-critical code, sure, write slop ahead. Critical code, read every line. \>\> So, two folks, same conference, day after day, talking about the one anxiety that we all feel. Should we all still be reading code in 2026? By the way, these two folks are the number six and number seven most watched YouTube videos from AI engineer from all time. So, they're obviously representing something that we're feeling we're talking about. And this is being the leadership track, something that you folks that report to you are talking about, okay? Should they still be reading code? And what's the what's the level quality? So, they name the same anxiety from both ends. Uh at this point, I probably should introduce Uh hey, I'm Alex Volkov. I'm the host of Thursday AI podcast. It's a podcast and a newsletter. We go live every week to talk about AI. For the past 3 and 1/2 years, we've been tracking every change in the AI engineering, every release from every lab, every model. And I'm also an AI evangelist with Weights & Biases and Core Weave. Um what also should I tell you about myself? That I've been covering AI engineers specifically since the first one in 2023. And oh boy, has it changed. And so, you can treat this as a dispatch from the front line because all of these people now are my friends. And we constantly talk about this in the speakers' room in the hallway track. I couldn't stop thinking about that tension. I couldn't stop thinking about that kind of disparity between the two folks, okay? And I put them both on a line, Zeshner from one end, Lapopolo on the other end. I called it a continuum. And I basically started asking people, "Hey, where are you on this line? Where are you a Zeshner? Do you still read every line of code? Are you a Lapopolo? Do you just YOLO and don't even look at code and think agents are good enough, etc.?" And I I got the framing wrong. But, I'll tell you about this in just a second, okay? So, before this, I want you to be kind of honest with yourself. And again, if you don't write code, or let me say this, if you don't babysit your own agents, but you you have reports that babysit agents for you, uh think about them when you answer this, okay? And be honest. On the ZL continuum, where are you? And let's take um let's take by vote of hands, who here has committed code that they've never looked at before? Amazing, love that. Who here still reads every line of code or at least critical code? I see one cowboy over there. I love that, man. I'm going to talk to you afterwards, okay? I want to understand exactly why you do this. Um, and so who's right? Let's talk about who's right. Let's talk about where we are right now, and we start with Ryan Lepopolo. If you get to meet Ryan over here, he's very AGI built. I think even within OpenAI, the AGI organization, Ryan is kind of like the more AGI built person. If you have a chance to go downstairs and grab the AGI pills that 2X prescribed, I think Ryan had all of them. He works at OpenAI, where he sits code is literally free. So are tokens. It's really We renamed Ryan, do you guys know the dash dash YOLO in Codex? It's kind of like the skip dangerous permissions in cloud code. So we renamed the Ryan Lepopolo YOLO Popolo. He's okay with it, by the way, I asked him. So if we check his kind of side, the the folks like him against the data, they're actually right. The optimists are right, at least about output. This is from Faros AI, I think I'm not the only speaker at this conference who sites this essay. It's sorry, this survey. It's new from April 2026. I think it's one of the best kind of evidence of where we're going that we can now site, okay? 22,000 engineers were surveyed about code. They call this the acceleration whiplash. And they're talking about my favorite stat on here, and you can read this yourself, 861% increase in code deletion per PR. So us together with AI agents, we love deleting code. Anthropic also said that they're shipping eight times more code per quarter than in 2025. But is it all good code? Okay, let's play a game. If you know the answer, you let me have my moment on here on stage, okay? But if you don't know the answer, let's guess. Whose status page is this? I think I I hear a few answers. I think most of us guessed it. This is Claude. Uh in fact, as you can see on the on the right, it was down when I took the screenshot. It was really funny. Uh this Anthropic is the company that probably uses the most AI-generated code, and their status page looks like a Christmas tree. Now, I'm not here to dunk on Anthropic. Uh they did an incredible job back on stage uh talking about Claude and etc. Um this may be due to scale, this may be due to other factors. Uh I'm not here to dunk on them, but \[snorts\] it just goes to show that they're not the only ones like this. Obviously, GitHub is famously also suffers from a little bit of golf. Output does not mean stability, okay? So, maybe this is a good example of what? Same essay, 31% increase in PRs merged with no review at all, human or agentic. Don't do this. I beg of you. Don't It's it's We'll talk about how to fix this in a second. So, when you ship this fast and this much, something gives, and usually it's quality. So, maybe Mario's right, yeah? Maybe the bill does come due in production. Same study, 242% increase in incident per PR. This kind of scary. The second study is also scary. Bugs per developer is up six times than 2025. So, even Anthropic can see this. I don't know if you guys read the RSI essay they posted, the recursive self-improvement, where they talk about, "Hey, what does the future hold?" They outline two scenarios. One of them says, "Maybe the acceleration will stop, and we're going to get used to this." They all They say, "That's actually not likely to happen. We just added this uh eventuality for clarity. We don't think that's likely to happen. What we think is going to happen is uh engineers and companies 10Xing to 100x-ing to 1,000x-ing their output and productivity, and then they say this, "We, as we began to push more code around the organization, human code review has become a new bottleneck." They're citing Amdahl's law that shows that if you have an explosion of productivity in one area, another area going to going to get blocked. And nobody removes the human in these organizations. In fact, careers in Anthropic and careers in OpenAI, they're still hiring humans. So, nobody's removing the human, and they're both saying that human code review is still a concern. And here's my mea culpa. I promise you I'll tell you where I got it wrong the framing. My mea culpa is the continuum is real. The Zeal continuum is real, but it's not about the people. It's about the tasks. The continuum is real. It's not about the people. It's about the task. Same engineer could be a Ryan LeCompte on Apollo on one piece of code, and has to be Mario Zechner and read every line of other pieces of code. Different task just need different proof. If we look at them closely, I obviously character- characterized them. I practiced this word multiple times and I still got it wrong. Characterized them. Their character on both ends for the Zeal continuum. But, if you look at them closely and what they're saying closely, they're actually not that different. Ryan's mechanism is moving attention up the layer. He's saying humans are unreliable at catching repeated mistakes of the same type. Repeatedly catching the mistakes of the same type. So, when they you do catch a mistake during the PR review, write the documentation, the linter, and the reviewer needs to remember this once, so the system will catch this type of bugs. He's not saying don't inspect your code. He's saying inspect the system, not every line. Mario from the other end is saying route by task. If it's not critical, let it rip. He said it. And if it's critical, you read every line. How do you know what's critical? Well, his answer is easy. You read the effing code. Uh my answer to add to this is also you ask your clankers. They're great at looking at a large repository and saying you and telling you, "Hey, this line is actually critical. You should look at this area. This primitives over here are critical." So, you ask your clanker. So, they agree more than I can have given credit for. And so, I think at the beginning of this, the wrong question is should I still be reading code in 2026? I think the better question right now for all of us is what proof does this specific change need? What proof does this specific change need? And so, I took Mario on the left, obviously. I took Ryan on the right. And then I took a bunch of other uh great AI engineers, friends, some friends, uh many of them speakers at this conference. I kind of distilled their advice down to a routing table. And uh they told me, I think Swyx told me on Twitter, "Just give me one slide that I will that people need to take a screenshot of. It's going to be this slide. You don't have to read it with me, but in the end, you're welcome to take a picture of this." This is the your Monday artifact. Routing the change where the proof needs it. Routing the change to the proof that it needs. You read every line of authentication, money movement, permissions, and irreversible data. You inspect the critical path yourself, and then obviously, you keep going. Uh decomposing, I think is very important. The more code is getting written, the more it's hard. Your eyes are starting to glaze over a a very long pull request. So, split it into atomic reviewable PRs. You know who's good at it? Agents. They're great at decomposing code. Ask them to do it. You verify that doesn't go away. This has been with us in the in engineering software engineering and the AI engineering doesn't go away. Traces, evals, shadow mode. Come talk to me after after this talk. I don't have enough time, but shadow mode is a really cool one that I learned about preparing this talk. And then, I think the most important one is separating. Many people have the same agent that writes the code, also inspects the outputs and writes the tests. Separating is very important. If you don't separate, it's kind of like if I came up with an exam and then I took an exam and I scored myself on the exam. It's not not really productive, right? And then last one is engineer. Rails observability rollback. This is what Ryan Le Popo talks about. Build the system that builds the system because read spends your attention once, engineer makes the system remember. Right? And you might sit in might be sitting there and saying, "Hey, did you hear the news, Alex? Fable is back. What about Fable? What about Mythos? Is this still relevant at this next scale of capability?" Because when I coined the Zeal Continuum, it was only 82 days ago. Mythos has just been announced. We weren't sure like what was going on. Only the people in Anthropic got access to it. And Jared S. Chipar, that was on stage from Anthropic, he said about Mythos and and Fable, "We used to check if Claude is doing the work right? And with Fable 5, I instead check if Claude is doing the right work." Let it land for a second. I don't know if you read the statement. When I read the statement, I felt like little chills at the back of my neck about the next like level of capability, okay? We used to check if Claude is doing the work right. With Fable, we check if Claude is doing the right work. And our favorite senpai who recently joined Anthropic and is getting unnecessary heat on Twitter said this. Andrej Karpathy has said, "It's never felt so tempting to stop looking at code at all. But don't do this in production." Senpai is great for the sole reason. Do you guys know the sentence, "This meeting could have been an email?" So, this presentation could have been Andrej Karpathy's one sentence, okay? He's naming the anxiety from both ends. It's never been so tempting to stop looking at code. Don't do this in production, even with Fable. And so, if you guys noticed, I have a little thing here. This. Uh it's so wide you can see my little laser pointer. Do you guys see the arrow, the capability drift arrow? This thing? When I wrote the continuum, I realized that it it's only a temporary place in time. Capability increases move us towards the popular. So, we're going to talk about capability increases as well because the review layer moves. If yesterday we inspected the outputs and we read the code, and today we inspect the task direction and kind of like directed to the right proof, maybe tomorrow we're inspecting the loops. Capability drift changes where proof belongs. It doesn't remove the requirement of proof. Talking about loops. Is that the next primitive? I think most of this conference, I think the zeitgeist for this one is going to be is talking factories and co-factories are real, and is loops is a real thing that I need to be doing at this point. By raising of hands, who here heard of loops? Keep your hands up, please, and take them down if you are not running loops right now and you haven't anything about they are. There's a good perception There's good number of people here who heard about loops, and they started with both these folks, Peter Steinberger, creator of Open Cloud Open Claw, and now is Open AI, and Boris Cherny. And pretty much within the span of 2 days, both of them start talking about loops that became kind of the zeitgeist. And loops are moving us from prompting each turn to designing the system that writes the actual prompts. By the way, do you guys know what's common between these guys and what's different between me and these guys? Their tokens are free. So, when they talk about loops and their tokens are free, they're not telling you, "Hey, you should be doing this right now specifically." But because they work at bigger labs, you can treat them as kind of a lighthouse that's pointing where we're all going. Kind of like Gretzky skate where the puck is going to to They're going to tell us what all of our enterprises are going to get get up on. And if it's if it's loops, then let me at least give you a TLDR, okay? Loops are basically fancy cron jobs that run on a schedule. But what they do is they discover a task and kind of start writing a prompt for this task from the plan. They write the plan, they execute, and most importantly for my talk here, they verify themselves, and if it doesn't work, they try again. So, an agent that loops grades its own work against the goal with less human intervention. But if the builder grades itself, you didn't remove the review, you hid it. Okay, this this connects to my routing table. This comes from Adi Olsmanyu recently at Google. He's also at this conference, a great engineer. Uh he said if I if I if I wasn't reviewing the code myself, or relied entirely on automated loops to fix my code, let's say a bug comes up in Jira and my loop picks it up and starts fixing this, my product quality would suffer. I'd likely end up in a downward spiral, digging myself into a deeper hole. So again, loops don't remove judgment, but they do raise the stakes on where you put it. So what about the future, folks? Nobody knows. Anthropic did not know that Claude code is going to explode on them and this is going to be a billion-dollar product. Uh nobody knew that coding agents and Harness are going to be the generalized agent, and now everybody's pursuing them. Folks at OpenAI with Codex, Elon with with Grok code, uh Google with with anti-gravity. Model capability is jumping at an insane pace, and what I implore and to tell you here is that flexibility is required. You need to keep being able to keep up with with the trends. This is why you stay engineer. And by the way, I told some folks here about my podcast Thursday and news. If you want to keep tracking where that line moves, feel free to scan this QR code, join our uh our, you know, newsletter, etc. Um and I'll leave you with this. Because it's my time, I'll leave you with this. Not every line in 2026 needs your eyes. Every system still needs your judgment. Thank you. \>\> \[applause\]