ah it was a bug in our codegen
November 4, 2025 at 3:06 AM
ah it was a bug in our codegen
@bnewbold.net any chance upcoming Go lexicon codegen changes will involve go-dasl?
November 5, 2025 at 6:46 PM
@bnewbold.net any chance upcoming Go lexicon codegen changes will involve go-dasl?
Been building on the Atmosphere? The official Lexicon codegen tooling has some rough edges for third-party devs:
• Bundle bloat
• Painful maintainability
• Inflexible client/server modes
Wrote up the pain points (and there might be a solution coming 👀)
• Bundle bloat
• Painful maintainability
• Inflexible client/server modes
Wrote up the pain points (and there might be a solution coming 👀)
Rethinking Lexicon Tooling for Third-Party Developers
The official Lexicon code generation tooling hasn't evolved much since its initial release. While it serves the core atproto codebase well, third-party developers face significant challenges with bundle size, manual maintenance, and architectural constraints that make building production-ready applications unnecessarily difficult.
matthieu.leaflet.pub
November 3, 2025 at 11:31 AM
Been building on the Atmosphere? The official Lexicon codegen tooling has some rough edges for third-party devs:
• Bundle bloat
• Painful maintainability
• Inflexible client/server modes
Wrote up the pain points (and there might be a solution coming 👀)
• Bundle bloat
• Painful maintainability
• Inflexible client/server modes
Wrote up the pain points (and there might be a solution coming 👀)
Spent the last 4 months looking into LLM codegen quality and techniques of improving it for the new APIs (not in the training set).
Many learnings from the process, with the key one being: LLMs "preffer" good API design.
Codegen fails? Your human devs would be likelly confused as well!
Many learnings from the process, with the key one being: LLMs "preffer" good API design.
Codegen fails? Your human devs would be likelly confused as well!
November 7, 2025 at 5:57 PM
Spent the last 4 months looking into LLM codegen quality and techniques of improving it for the new APIs (not in the training set).
Many learnings from the process, with the key one being: LLMs "preffer" good API design.
Codegen fails? Your human devs would be likelly confused as well!
Many learnings from the process, with the key one being: LLMs "preffer" good API design.
Codegen fails? Your human devs would be likelly confused as well!
@garrytan https://x.com/garrytan/status/1986176749139591446 #x-garrytan
Quote: https://x.com/venturetwins/status/1986136181302829156
Wabi is one of the most important new consumer codegen apps — it's a little bit of an Instagram moment for true no-code all on mobile.
A...
Quote: https://x.com/venturetwins/status/1986136181302829156
Wabi is one of the most important new consumer codegen apps — it's a little bit of an Instagram moment for true no-code all on mobile.
A...
November 5, 2025 at 9:16 PM
@garrytan https://x.com/garrytan/status/1986176749139591446 #x-garrytan
Quote: https://x.com/venturetwins/status/1986136181302829156
Wabi is one of the most important new consumer codegen apps — it's a little bit of an Instagram moment for true no-code all on mobile.
A...
Quote: https://x.com/venturetwins/status/1986136181302829156
Wabi is one of the most important new consumer codegen apps — it's a little bit of an Instagram moment for true no-code all on mobile.
A...
Stop coding the hard way! The future of Angular is agentic.
Use Gemini Canvas for rapid prototyping. Leverage angular.dev/ai for advanced prompts & context. And check your LLM's output quality with the open-source Web Codegen Scorer!
AI isn't replacing developers; it's empowering them. #Angular
Use Gemini Canvas for rapid prototyping. Leverage angular.dev/ai for advanced prompts & context. And check your LLM's output quality with the open-source Web Codegen Scorer!
AI isn't replacing developers; it's empowering them. #Angular
Get Started • Angular
The web development framework for building modern apps.
angular.dev
November 5, 2025 at 7:09 PM
Stop coding the hard way! The future of Angular is agentic.
Use Gemini Canvas for rapid prototyping. Leverage angular.dev/ai for advanced prompts & context. And check your LLM's output quality with the open-source Web Codegen Scorer!
AI isn't replacing developers; it's empowering them. #Angular
Use Gemini Canvas for rapid prototyping. Leverage angular.dev/ai for advanced prompts & context. And check your LLM's output quality with the open-source Web Codegen Scorer!
AI isn't replacing developers; it's empowering them. #Angular
LLVM noalias about to mutilate the codegen on this
November 1, 2025 at 4:21 AM
LLVM noalias about to mutilate the codegen on this
When coding with AI, I follow a fairly simple principle:
If prompting effort > coding effort, code it. 😅
Would it be useful if I made a video with examples, etc?
If prompting effort > coding effort, code it. 😅
Would it be useful if I made a video with examples, etc?
November 6, 2025 at 9:21 AM
When coding with AI, I follow a fairly simple principle:
If prompting effort > coding effort, code it. 😅
Would it be useful if I made a video with examples, etc?
If prompting effort > coding effort, code it. 😅
Would it be useful if I made a video with examples, etc?
Second time I heard this pitch this week. The problem is that (well-maintained) dependencies take care of the issues you haven’t thought of yet. There is some cost to having dependencies and making sure the dependency graph doesn’t grow too much takes work, but codegen doesn’t seem a great option.
What if, instead of adding dependencies for most things, we used large language models to generate code instead?
nrempel.com/smaller-sur...
nrempel.com/smaller-sur...
Smaller Surfaces
Modern dependency graphs feel like thickets. Agents make it practical to generate narrow helpers, keep the runtime surface small, and verify the results with the tools you already trust. This essay looks at how to use generation over integration without slipping into wheel reinvention or fragile bespoke stacks.
nrempel.com
November 1, 2025 at 6:25 PM
Second time I heard this pitch this week. The problem is that (well-maintained) dependencies take care of the issues you haven’t thought of yet. There is some cost to having dependencies and making sure the dependency graph doesn’t grow too much takes work, but codegen doesn’t seem a great option.
> Parsing: 150 msec
> Type Checking: 300 msec
> Borrow Checking: 200 msec
> Optimization Passes: 800 msec
> Codegen: 400 msec
> Linking: 300 msec
> Proc Macros: 7 min
Somebody who's good at compilers please help, my build time is dying
> Type Checking: 300 msec
> Borrow Checking: 200 msec
> Optimization Passes: 800 msec
> Codegen: 400 msec
> Linking: 300 msec
> Proc Macros: 7 min
Somebody who's good at compilers please help, my build time is dying
November 3, 2025 at 5:32 PM
> Parsing: 150 msec
> Type Checking: 300 msec
> Borrow Checking: 200 msec
> Optimization Passes: 800 msec
> Codegen: 400 msec
> Linking: 300 msec
> Proc Macros: 7 min
Somebody who's good at compilers please help, my build time is dying
> Type Checking: 300 msec
> Borrow Checking: 200 msec
> Optimization Passes: 800 msec
> Codegen: 400 msec
> Linking: 300 msec
> Proc Macros: 7 min
Somebody who's good at compilers please help, my build time is dying
Thank you Svelte team for adding ts-estree codegen support to esrap.
Now you can parse and print code in node.js via oxc-parser and esrap.
github.com/sveltejs/es...
Now you can parse and print code in node.js via oxc-parser and esrap.
github.com/sveltejs/es...
Release v2.1.1 · sveltejs/esrap
Patch Changes
9a7693e: fix: support more typescript nodes
github.com
October 29, 2025 at 1:36 PM
Thank you Svelte team for adding ts-estree codegen support to esrap.
Now you can parse and print code in node.js via oxc-parser and esrap.
github.com/sveltejs/es...
Now you can parse and print code in node.js via oxc-parser and esrap.
github.com/sveltejs/es...
Funny was just investigating the same. They have a docs page for it actually - playwright.dev/docs/codegen
Test generator | Playwright
Introduction
playwright.dev
October 29, 2025 at 2:30 PM
Funny was just investigating the same. They have a docs page for it actually - playwright.dev/docs/codegen
the lexicon codegen in there i deleted was pretty big. probably 10-12k lines overall, if i were to do the math was codegen that's now basically just in jacquard-api. but there was a TON of boilerplate i'd written around atrium i could just strip.
October 29, 2025 at 12:03 AM
the lexicon codegen in there i deleted was pretty big. probably 10-12k lines overall, if i were to do the math was codegen that's now basically just in jacquard-api. but there was a TON of boilerplate i'd written around atrium i could just strip.
phew. all the bon builder derives in jacquard-api were getting problematic for compile times on my machine (since i need to check with all features regularly to ensure codegen didn't break), so i extracted the bits of bon i needed into the api code generation stage, to pay the syn tax up front.
October 27, 2025 at 9:18 PM
phew. all the bon builder derives in jacquard-api were getting problematic for compile times on my machine (since i need to check with all features regularly to ensure codegen didn't break), so i extracted the bits of bon i needed into the api code generation stage, to pay the syn tax up front.
#GoogleAIStudio Does #Angular-"AI Studio can now generate Angular applications. Add the Web Codegen Scorer tool on top and you get a pretty solid Angular development platform".
On IProgrammer: cutt.ly/Ir7pznzI
@angular.dev
On IProgrammer: cutt.ly/Ir7pznzI
@angular.dev
October 27, 2025 at 7:08 PM
#GoogleAIStudio Does #Angular-"AI Studio can now generate Angular applications. Add the Web Codegen Scorer tool on top and you get a pretty solid Angular development platform".
On IProgrammer: cutt.ly/Ir7pznzI
@angular.dev
On IProgrammer: cutt.ly/Ir7pznzI
@angular.dev
graphql.tada is amazing.
I faffed around with various codegen settings and different vscode plugins and they couldn’t seem to agree on what constitutes a query.
But graphql.tada just works, with no watchers, no scripts, no language server.
I faffed around with various codegen settings and different vscode plugins and they couldn’t seem to agree on what constitutes a query.
But graphql.tada just works, with no watchers, no scripts, no language server.
October 24, 2025 at 3:53 PM
graphql.tada is amazing.
I faffed around with various codegen settings and different vscode plugins and they couldn’t seem to agree on what constitutes a query.
But graphql.tada just works, with no watchers, no scripts, no language server.
I faffed around with various codegen settings and different vscode plugins and they couldn’t seem to agree on what constitutes a query.
But graphql.tada just works, with no watchers, no scripts, no language server.
in my understanding, it's because having an explicit ⊥ is used by rustc to make good codegen. being able to distinguish divergence from universal quantification is helpful for stuff like dead code elimination, borrow checking, etc.
October 21, 2025 at 4:15 PM
in my understanding, it's because having an explicit ⊥ is used by rustc to make good codegen. being able to distinguish divergence from universal quantification is helpful for stuff like dead code elimination, borrow checking, etc.
Has there ever been a truly critical memory safety vulnerability in widely used Rust code?
Excluding codegen bugs (like the one in wasmtime), since unfortunately the safety guarantees of Rust only extend to actual rust code, not code generated by rust code.
Excluding codegen bugs (like the one in wasmtime), since unfortunately the safety guarantees of Rust only extend to actual rust code, not code generated by rust code.
October 21, 2025 at 5:09 PM
Has there ever been a truly critical memory safety vulnerability in widely used Rust code?
Excluding codegen bugs (like the one in wasmtime), since unfortunately the safety guarantees of Rust only extend to actual rust code, not code generated by rust code.
Excluding codegen bugs (like the one in wasmtime), since unfortunately the safety guarantees of Rust only extend to actual rust code, not code generated by rust code.
@ ts developers can you make it so that the codegen emits `typeof X` instead of expanding them
October 21, 2025 at 5:41 AM
@ ts developers can you make it so that the codegen emits `typeof X` instead of expanding them
Some developers are just fascinated by codegen. It’s like the Holy Grail to them. I once replaced a consultant who’d spent over a year writing software that would generate SQL queries on the fly based on a form that insurance underwriters would fill out.
…
…
October 20, 2025 at 4:04 AM
Some developers are just fascinated by codegen. It’s like the Holy Grail to them. I once replaced a consultant who’d spent over a year writing software that would generate SQL queries on the fly based on a form that insurance underwriters would fill out.
…
…
The one I haven’t published yet 🥲
But it’s a strongly typed no-codegen ORM that has big Drizzle vibes
But it’s a strongly typed no-codegen ORM that has big Drizzle vibes
October 20, 2025 at 6:31 AM
The one I haven’t published yet 🥲
But it’s a strongly typed no-codegen ORM that has big Drizzle vibes
But it’s a strongly typed no-codegen ORM that has big Drizzle vibes
at this point, i think i prefer `go generate` over build dot rs scripts. the latter are too annoying to parallelize and i think codegen is rare enough that being able to run it as an explicit step is worth it for faster/more reliable IDEs
October 18, 2025 at 4:15 PM
at this point, i think i prefer `go generate` over build dot rs scripts. the latter are too annoying to parallelize and i think codegen is rare enough that being able to run it as an explicit step is worth it for faster/more reliable IDEs
AI tools still struggle with accessible code.
Aaron Gustafson digs into where CodeGen models miss the mark and what we can do about it.
www.aaron-gustafson.com/notebook/ide...
#Accessibility #AI #CodeGen #InclusiveTech #A11y #WebDev #DigitalAccessibility
Aaron Gustafson digs into where CodeGen models miss the mark and what we can do about it.
www.aaron-gustafson.com/notebook/ide...
#Accessibility #AI #CodeGen #InclusiveTech #A11y #WebDev #DigitalAccessibility
Identifying Accessibility Data Gaps in CodeGen Models :: Aaron Gustafson
Late last year, I probed an LLM’s responses to HTML code generation prompts to assess its adherence to accessibility best practices. The results were unsurprisingly disappointing — roughly what I’d…
www.aaron-gustafson.com
October 18, 2025 at 12:30 PM
AI tools still struggle with accessible code.
Aaron Gustafson digs into where CodeGen models miss the mark and what we can do about it.
www.aaron-gustafson.com/notebook/ide...
#Accessibility #AI #CodeGen #InclusiveTech #A11y #WebDev #DigitalAccessibility
Aaron Gustafson digs into where CodeGen models miss the mark and what we can do about it.
www.aaron-gustafson.com/notebook/ide...
#Accessibility #AI #CodeGen #InclusiveTech #A11y #WebDev #DigitalAccessibility
Hardest question, not sure I have a single take. Actual valuable use cases of LLMs seem to be consolidating around codegen, MCP and generally tech selling to tech.
Video, deepfakes, and cognitive offloading, especially in education just feels downright depressing to me.
Video, deepfakes, and cognitive offloading, especially in education just feels downright depressing to me.
October 18, 2025 at 4:04 AM
Hardest question, not sure I have a single take. Actual valuable use cases of LLMs seem to be consolidating around codegen, MCP and generally tech selling to tech.
Video, deepfakes, and cognitive offloading, especially in education just feels downright depressing to me.
Video, deepfakes, and cognitive offloading, especially in education just feels downright depressing to me.