Video · 2026-07-06 · 1h 29m · 24 moments

Do You Read The Code? ⟡ HTTP Gets a NEW Method ⟡ Codex Gets Sol Ultra ⌁ Syntax Weekly ⌁

✦ AI generated

GUEST David Cramer GUEST Dax GUEST CJ Reynolds GUEST Scott Tolinski

timeline · colored by role

Claim◆

Whether you read AI-generated code should depend entirely on the project: for personal projects and one-off scripts it's fine to auto-accept edits without reading, but for production code you ship to a company or contribute to, you should read it as an engineer.

Scott Tolinski argues the 'do you read the code' debate isn't binary — it depends on whether the project is a disposable personal script or real production code you're shipping.

transcript

Scott Tolinski: if you've seen my takes on this channel, you would expect that yes, I read the code. But I will say for some personal projects, for some one-off scripts, I don't. I do actually turn on auto accept edits... but it when it comes to production code, when it comes to like code that you're shipping for your company or that you're contributing to, you probably should read the code just as an engineer.

rebuts · 1

Claim◆

Whether you read AI-generated code should depend on the situation: skip it for disposable personal projects and one-off scripts, but read it carefully for production code or anything shipped to a company or contributed to a team.

Scott argues reading AI-generated code isn't a universal rule — it depends on whether the project matters (production/shared work) or is throwaway (personal scripts).

transcript

Scott Tolinski: if you've seen my takes on this channel, you would expect that yes, I read the code. But I will say for some personal projects, for some one-off scripts, I don't. I do actually turn on auto accept edits. And that's because a lot of times those aren't projects that are going to have a long life. It's something that I'm I'm just testing out. I want to get it up and going.

extends · 1supports · 1

Claim

No good engineer will ever tell you that reading your own code is useless, and claiming otherwise is itself a reliable signal of poor engineering judgment.

Quoting Sentry's David Cramer, Scott says any engineer who dismisses reading code as pointless is revealing a red flag about their own engineering discipline.

transcript

David Cramer: No good engineer will tell you it's useless to read the code. By the way, that's the easiest signal. I think like it's part of our jobs like we have to care. I think like if you're too lost in the sauce if you just think that just prompt prompt prompt prompt prompt.

Claim

Even on a single team, every developer ends up with their own individual AI workflow, and the industry hasn't figured out how to collaborate on those workflows together — like sharing skill files — the way real multiplayer software engineering requires.

CJ Reynolds says that despite AI hype claiming solo devs with agents can replace teams, nobody has actually solved how teams share and collaborate on agent workflows — software is still fundamentally a multiplayer endeavor.

transcript

CJ Reynolds: even though I'm on the syntax team I'm still kind of a solo dev in the sense that I'm working on on a lot of side projects and all the workflows that I've created are just for me and my ability to write code and review it but I haven't even thought about how that works on a team and I think that's going to be the answer moving forward.

extends · 2provides context · 1

Claim

People who claim reading code no longer matters should be challenged to prove it with results — if they're really so far ahead by not reading the code, they should be visibly richer or more successful because of it.

CJ relays a tweet from Dax arguing that people dismissing the value of reading code should have to show tangible proof (money, productivity) of their approach's superiority, implying most don't.

transcript

Dax: Guys, stop trying to argue with the code doesn't matter posts and just ask them if they're so ahead of the game, why aren't they richer? Like he has multiple posts like this. Like if if if really we don't have to read the code anymore code anymore, like show me the receipts.

supports · 1

Claim◆

Whether you read AI-generated code should depend on whether it's a project you actually care about — for personal projects and one-off scripts it's fine to skip it, but for production code you should read it, because working code isn't necessarily good code.

Scott argues there's no universal answer to the 'should you read AI-written code' debate; it depends on whether the project matters, with production code demanding review since something working doesn't mean it's good.

transcript

Scott Tolinski: I think the most exhausting answer and the correct answer to this is it depends if it's something you care about it's absolutely worth still reading the code and if the question that well if it works does it matter? Well just because something works doesn't mean it's good. Lots of works that's no good.

extends · 1

Claim◆

Whether you should read AI-generated code is entirely situational — if it's a project you care about or production code, you must read it, but for throwaway personal scripts it's fine to auto-accept; and 'it works' is never sufficient justification for not reading code, since working code isn't necessarily good code.

Scott argues the 'should you read the code' debate has a nuanced, situational answer: read it for projects/production code you care about, skip it for disposable scripts, and never accept 'it works' as proof the code is good.

transcript

Scott Tolinski: the most exhausting answer and the correct answer to this is it depends if it's something you care about it's absolutely worth still reading the code and if the question that well if it works does it matter? Well just because something works doesn't mean it's good. Lots of works that's no good. So works is not the bar and works has never been the bar.

supports · 1

Mechanism

HTTP QUERY is a new proposed method that behaves like an idempotent GET but can carry a body (form-encoded or JSON), replacing the common hack of stuffing data into GET request bodies or misusing POST for read-only requests like GraphQL queries.

CJ explains the new HTTP QUERY method (RFC), currently implemented in Node.js's HTTP module, which lets developers send GET-like idempotent requests with a body instead of hacking GET or misusing POST.

transcript

CJ Reynolds: Essentially, it's an HTTP get, but it can have a body. So you can put those query params in the body as like form URL encoded params, or you could have JSON inside of the body. You could set the content type to whatever you want. But the idea is this will still be in classified as an item potent request.

Mechanism

The new HTTP QUERY method (RFC-based) is essentially a GET request that is allowed to carry a body, letting complex query data live in the request body instead of being crammed into a URL query string, while still being classified as an idempotent, non-modifying request.

CJ explains the new proposed HTTP QUERY method, currently implemented in Node.js's HTTP module, which solves the problem of needing a GET-like idempotent request that can carry a JSON or form-encoded body (as GraphQL APIs currently fake using POST).

transcript

CJ Reynolds: Essentially, it's an HTTP get, but it can have a body. So you can put those query params in the body as like form URL encoded params, or you could have JSON inside of the body... But the idea is this will still be classified as an item potent request. It doesn't modify anything on the server. It's really just requesting data.

Mechanism

The new HTTP QUERY method works like GET but allows a request body, letting idempotent read requests carry complex query data (as form-encoded or JSON) instead of hacking it into POST requests the way GraphQL APIs currently do.

CJ explains the newly proposed HTTP QUERY method, an idempotent GET-like request that can carry a body, aimed at replacing the common hack of using POST for read-only requests like GraphQL queries.

transcript

CJ Reynolds: Essentially, it's an HTTP get, but it can have a body. So you can put those query params in the body as like form URL encoded params, or you could have JSON inside of the body... the idea is this will still be in classified as an item potent request. It doesn't modify anything on the server. It's really just requesting data.

Mechanism

The proposed HTTP QUERY method (RFC 10008) functions like GET but allows a request body, giving developers an official, idempotent way to send complex query data instead of misusing POST or hacking a body onto GET.

CJ explains a new proposed HTTP method called QUERY, already implemented in Node.js's HTTP module, that lets developers send an idempotent, GET-like request with a JSON or form-encoded body — solving the common workaround of using POST (as GraphQL does) just to carry query data.

transcript

Claim◆

The outrage over GitHub's joke CD-ROM promotion is absurd because a small marketing team offering a novelty item has essentially zero effect on GitHub's actual infrastructure or uptime.

Scott mocks the backlash to GitHub's tongue-in-cheek CD-ROM promotion (riffing on PlayStation going discless), arguing critics wrongly blame a small marketing team for the company's infrastructure and uptime issues.

transcript

Scott Tolinski: Why don't you fix your infrastructure? as if the marketing team at GitHub, who's probably just a couple of people putting this promotion on, are the ones going to be single-handedly fixing your infrastructure, like, oh, let's let's redivert that energy and money.

Context

The backlash claiming GitHub's fun CD-ROM promotion diverts money or effort from fixing GitHub's infrastructure/uptime problems is out of touch, since a small marketing team producing a joke promotion has no bearing on infrastructure spending.

Scott rants that critics blaming GitHub's joke CD-ROM promotion for its infrastructure/uptime issues fundamentally misunderstand how large companies and marketing teams operate.

transcript

Scott Tolinski: the people who are saying like it's diverting funds from infrastructure or this is why GitHub is down all the time like you people are so out to lunch uh and and and I don't know I got somebody grumpy at me I'm gonna I'm put them on blast actually let me even let me let me do that

Claim

React's hook model is flawed because all the code inside a function component's hooks re-runs on every single render, whereas frameworks like SolidJS and the new Gia framework fix this by using class-based components whose bodies run once, with methods and effects handling later updates.

CJ contrasts React's re-render-everything hooks model with the new Gia framework (and SolidJS before it), which uses class components that run once and expose state/effects as methods, calling this a better mental model for reasoning about code.

transcript

CJ Reynolds: The every hook that you write inside of a component is run on every single render. Not necessarily invoked, but it's recreated on every render... And I don't like that. I think it's a very silly model. And SolidJS solved this where essentially you write your SolidJS component, it runs once and then you have effects that can potentially run later... And GIA does that as well.

Claim

React's hooks model is flawed because every hook inside a function component gets re-run on every single render, whereas frameworks like SolidJS and the new Gia framework run components once and use class-based state and methods to avoid that overhead.

CJ criticizes React's hooks system for re-running all component code on every render, and praises the new Gia framework (like SolidJS before it) for running components once and using class-based state/methods instead.

transcript

CJ Reynolds: One of the things I don't like about React is the entire hooks system. The every hook that you write inside of a component is run on every single render. Not necessarily invoked, but it's recreated on every render... And I don't like that. I think it's a very silly model.

Claim

Gia's approach of putting component state into class properties and methods, rather than React's hooks that rerun on every render, provides a clearer and better mental model for managing interactive state.

CJ praises the new JS framework Gia for reviving class-based components for stateful UI, contrasting it favorably with React's hooks, which he criticizes as hard to reason about.

transcript

CJ Reynolds: if you're going to build a component that needs internal state, you actually use a class component. So it kind of goes back to like the old React way of uh class extends component and then instead of a render function you have a template function. But what that means is you can have class methods.

Anecdote◆

Fable is, on a vibes level, the best AI model Scott has used, standing out for reliably executing large, complex tasks with less handholding than other models.

Scott shares his hands-on experience with the Fable AI model before losing access, saying it excelled at executing large, complex tasks reliably and required less back-and-forth than other models he's tried.

transcript

Scott Tolinski: I'm here to tell you that on a vibes level, it's definitely the best model that I've used. It's done a whole lot of stuff. And one thing that I really liked was the ability for it to execute large things really reliably... it was able to do more complex things and it did so better, and with less handholding.

explains mechanism · 1supports · 1

Example

In a head-to-head planning test across Fable, GPT-5.5, Opus 4.8, and Kimi 2.7, Fable stood out as far more agentic — it went beyond the prompt to actively research and reverse-engineer an external API rather than just producing a plan, though at a much higher usage/credit cost.

CJ compares planning the same project across four models and finds Fable uniquely proactive, digging in to reverse-engineer an API on its own instead of just returning a plan, at the cost of burning through usage credits much faster.

transcript

CJ Reynolds: Fable just went just started researching. It was like okay this is what you want to build but I see you also want to talk to the Amazon API and so it started reverse engineering the Amazon music API and like it was just tool call after tool call. So it definitely is more agentic basically the way that they've trained it is it it will try and try and try more before it's done.

Fact

Anthropic quietly added a feature to Claude Code where, if a user doesn't respond to a clarifying question within 60 seconds (and the terminal isn't focused), it auto-selects an answer and keeps coding — a behavior that sparked backlash and multiple reports of unwanted production changes before it was made off by default.

CJ recounts a GitHub-reported bug/feature where Claude Code auto-picks an answer and continues coding if the user is unresponsive for 60 seconds while the terminal isn't focused, which upset many users and was eventually disabled by default after complaints.

transcript

CJ Reynolds: the Claude team secretly added this feature where if the user does not answer within 60 seconds, it just automatically chooses an answer and keeps going... There are multiple stories in the GitHub comments of like this happened to me like I was running three different sessions and I don't always have every session in focus. So it just kept on coding and then like did some stuff in production I didn't want it to.

Fact

Claude Code was secretly auto-selecting an answer to its own clarifying questions and continuing to code if the user didn't respond within 60 seconds, a default behavior many users discovered only after it made unwanted changes.

CJ describes a widely-criticized Claude Code bug/feature where the agent would auto-pick an answer and keep coding if the user didn't respond to a clarifying question within 60 seconds, which was eventually turned off by default after backlash.

transcript

CJ Reynolds: So maybe you're writing a plan. It'll ask you, oh do you want to use this versus that or how do we want do we want to approach it this way or that way? And you can answer that question with a prompt or by choosing from multiple choice. But the Claude team secretly added this feature where if the user does not answer within 60 seconds, it just automatically chooses an answer and keeps going.

Fact◆

Claude Code's CLI covertly fingerprints where a user is running from — using different Unicode characters (slashes instead of dashes, a different apostrophe) in its system prompt when it detects the user is in specific Chinese time zones — likely to help detect and deter distillation attacks by Chinese AI labs.

CJ describes a reverse-engineering finding that Claude Code subtly changes formatting in its system prompt (date format, apostrophe Unicode) when it detects the user is in Chinese time zones or domains, likely to detect attempts by rival Chinese labs to distill Claude's outputs.

transcript

CJ Reynolds: If it detects that you're in either of these two Chinese time zones, it changes the prompt to have slashes instead of dashes and then it changes the apostrophe to use a different unicode value instead... this is essentially to help prevent distillation attacks where some of these Chinese labs like DeepSeek and some of the others like Zai are potentially prompting Claude, saving the responses, and then using that as training for their own models.

rebuts · 1

Fact◆

Claude Code covertly alters the formatting of its system prompt (using slashes instead of dashes, and a different apostrophe unicode character) when it detects the user is in a Chinese time zone, effectively fingerprinting the request's origin, likely to help detect and prevent Chinese labs from distilling Claude's outputs into their own models.

CJ describes reverse-engineering findings showing Claude Code steganographically alters prompt formatting (date separators, apostrophe characters) when it detects Chinese time zones or domains linked to labs like DeepSeek, apparently to fingerprint usage and guard against distillation attacks.

transcript

CJ Reynolds: if it detects that you're in either of these two Chinese time zones, it changes the prompt to have slashes instead of dashes and then it changes the apostrophe to use a different unicode value instead. So if you were working on code in China... you would kind of have no idea that it basically just fingerprinted you as someone using cloud code in China.

Fact◆

The Claude Code CLI covertly alters the format of its system prompt (using slashes instead of dashes, and a different apostrophe unicode character) when it detects the user is in Chinese time zones like Shanghai or northwestern China, effectively fingerprinting Chinese users to Anthropic's backend, likely to help detect distillation attempts by Chinese AI labs.

CJ describes a reverse-engineering finding that Claude Code steganographically alters date formatting and punctuation in its prompts based on the user's Chinese time zone or domain, likely to detect and deter model-distillation attempts by rival Chinese AI labs.

transcript

CJ Reynolds: there is some code that will detect like if you're running cloud code from the Asia Shanghai time zone or the Asia... basically, if it detects that you're in either of these two Chinese time zones, it changes the prompt to have slashes instead of dashes and then it changes the apostrophe to use a different unicode value instead... it basically just fingerprinted you as someone using cloud code in China.

rebuts · 1

Fact◆

Claude Code secretly alters formatting details in its system prompt (using slashes instead of dashes and a different apostrophe unicode character) specifically when it detects the user is in a Chinese time zone, effectively fingerprinting Chinese users likely to help counter model-distillation attacks by Chinese AI labs.

CJ reports that reverse-engineering of Claude Code revealed it steganographically marks its prompts differently when run from Chinese time zones, likely to detect/deter distillation by Chinese AI labs.

transcript

Highlight slides

Reading AI Code Isn't All-or-Nothing✦ from: Whether you read AI-generated code should depend on the situation: skip it for disposable personal projects and one-off scripts, but read it carefully for production code or anything shipped to a company or contributed to a team.It Depends on the Project✦ from: Whether you read AI-generated code should depend entirely on the project: for personal projects and one-off scripts it's fine to auto-accept edits without reading, but for production code you ship to a company or contribute to, you should read it as an engineer.Personal Projects vs Production Code✦ from: Whether you read AI-generated code should depend entirely on the project: for personal projects and one-off scripts it's fine to auto-accept edits without reading, but for production code you ship to a company or contribute to, you should read it as an engineer.Scott's Own Practice✦ from: Whether you read AI-generated code should depend on the situation: skip it for disposable personal projects and one-off scripts, but read it carefully for production code or anything shipped to a company or contributed to a team.Should You Read AI-Generated Code? It Depends✦ from: Whether you read AI-generated code should depend on whether it's a project you actually care about — for personal projects and one-off scripts it's fine to skip it, but for production code you should read it, because working code isn't necessarily good code.Should You Read AI-Generated Code? It Depends✦ from: Whether you should read AI-generated code is entirely situational — if it's a project you care about or production code, you must read it, but for throwaway personal scripts it's fine to auto-accept; and 'it works' is never sufficient justification for not reading code, since working code isn't necessarily good code.Working Isn't the Same as Good✦ from: Whether you read AI-generated code should depend on whether it's a project you actually care about — for personal projects and one-off scripts it's fine to skip it, but for production code you should read it, because working code isn't necessarily good code.'It Works' Is Not the Bar✦ from: Whether you should read AI-generated code is entirely situational — if it's a project you care about or production code, you must read it, but for throwaway personal scripts it's fine to auto-accept; and 'it works' is never sufficient justification for not reading code, since working code isn't necessarily good code.'It Works' Is Not Enough✦ from: Whether you should read AI-generated code is entirely situational — if it's a project you care about or production code, you must read it, but for throwaway personal scripts it's fine to auto-accept; and 'it works' is never sufficient justification for not reading code, since working code isn't necessarily good code.Outrage Over GitHub's CD-ROM Joke Is Misplaced✦ from: The outrage over GitHub's joke CD-ROM promotion is absurd because a small marketing team offering a novelty item has essentially zero effect on GitHub's actual infrastructure or uptime.Scott's Point✦ from: The outrage over GitHub's joke CD-ROM promotion is absurd because a small marketing team offering a novelty item has essentially zero effect on GitHub's actual infrastructure or uptime.Fable: Best Model, On Vibes✦ from: Fable is, on a vibes level, the best AI model Scott has used, standing out for reliably executing large, complex tasks with less handholding than other models.Why It Stood Out✦ from: Fable is, on a vibes level, the best AI model Scott has used, standing out for reliably executing large, complex tasks with less handholding than other models.Claude Code Fingerprints Chinese Time Zones✦ from: Claude Code's CLI covertly fingerprints where a user is running from — using different Unicode characters (slashes instead of dashes, a different apostrophe) in its system prompt when it detects the user is in specific Chinese time zones — likely to help detect and deter distillation attacks by Chinese AI labs.Why: Deterring Distillation Attacks✦ from: Claude Code's CLI covertly fingerprints where a user is running from — using different Unicode characters (slashes instead of dashes, a different apostrophe) in its system prompt when it detects the user is in specific Chinese time zones — likely to help detect and deter distillation attacks by Chinese AI labs.Claude Code Fingerprints Chinese Time Zones✦ from: Claude Code covertly alters the formatting of its system prompt (using slashes instead of dashes, and a different apostrophe unicode character) when it detects the user is in a Chinese time zone, effectively fingerprinting the request's origin, likely to help detect and prevent Chinese labs from distilling Claude's outputs into their own models.Likely Purpose: Anti-Distillation Defense✦ from: Claude Code covertly alters the formatting of its system prompt (using slashes instead of dashes, and a different apostrophe unicode character) when it detects the user is in a Chinese time zone, effectively fingerprinting the request's origin, likely to help detect and prevent Chinese labs from distilling Claude's outputs into their own models.Claude Code Fingerprints Chinese Users✦ from: The Claude Code CLI covertly alters the format of its system prompt (using slashes instead of dashes, and a different apostrophe unicode character) when it detects the user is in Chinese time zones like Shanghai or northwestern China, effectively fingerprinting Chinese users to Anthropic's backend, likely to help detect distillation attempts by Chinese AI labs.Likely Purpose: Anti-Distillation Signal✦ from: The Claude Code CLI covertly alters the format of its system prompt (using slashes instead of dashes, and a different apostrophe unicode character) when it detects the user is in Chinese time zones like Shanghai or northwestern China, effectively fingerprinting Chinese users to Anthropic's backend, likely to help detect distillation attempts by Chinese AI labs.Claude Code Fingerprints Chinese Time Zones✦ from: Claude Code secretly alters formatting details in its system prompt (using slashes instead of dashes and a different apostrophe unicode character) specifically when it detects the user is in a Chinese time zone, effectively fingerprinting Chinese users likely to help counter model-distillation attacks by Chinese AI labs.Likely Motive: Countering Model Distillation✦ from: Claude Code secretly alters formatting details in its system prompt (using slashes instead of dashes and a different apostrophe unicode character) specifically when it detects the user is in a Chinese time zone, effectively fingerprinting Chinese users likely to help counter model-distillation attacks by Chinese AI labs.

Related episodes

Video · The Cognitive Revolution

Fable's Back, AI Engineer Recap, & SambaNova

→

Video · 20VC

Jensen's Open-Weights Letter | Google Cloud Grows 82% But The Market Tanks

→

Audio · All-In Podcast

Elon's Anthropic Deal, The Next AI Monopoly?, "FDA for AI" Panic, Trading the AI Boom

→