Code search in 2026 vs in the past.

Would love to see a score/eval of my generated search, though, to see if the AI missed an instantiation.

We stop debating *how* something is done and start scrutinizing *what* it should do. We become designers of intent.

This isn't about writing less code. It's about focusing on the right code. The contracts. The boundaries. The intent.

We'll work in "human files"—interfaces, behavior, constraints. Implementation gets pushed to *.ai.ts files, generated by AI.

A mental model for the future of dev work 🧵

You're absolutely right...

🤖 🤛

With enough tokens, all bugs are shallow.

I wonder what the future Github insights should show.

% of commits by ai, % of lines by ai, and so on?

Seems like that gamification is no longer as appealing in today's world.

Perhaps a new # of shipped features or value added that is inferred may be more appropriate.

Context switching hurts. If we can figure out a way in this AI world to improve our context switching. Would be profitable and sustainable

Would be nice to see github auto link submodule PRs from/to the monorepo PR. Github should have all the things to get this going automatically,

Happy filing day for CPAs. Remember to post pics of your hard earned holidays.

Prior to 2023 I relegated the usefulness of submodules. 2025, in the world of LLMs and large context windows, submodules are almost a must for effective pair programming with AI.

Braintrust - The evals and observability platform for building reliable AI agents

This is more than just a relocation—it's the start of a new chapter and a significant commitment to building at Braintrust (braintrust.dev).

I can't wait to collaborate with the brilliant minds in this ecosystem and prove why our platform is the best for AI development.

It's official: after 15+ years, I've made the move to the Bay Area! The transition was challenging but rewarding, and I'm deeply appreciative of this new opportunity.

Today, you can ask the LLM to be an expert and pair program with you.

Spend more time being a great QA and architectural engineer. Follow-up with "Why's" and "What else should I consider?"

What took me 15 years of experience may take you months to catch up. No excuses!

Vibe coding doesn't replace hard earned skills.

If you're early in your career use AI coding tools to shorten the feedback loop.

In the past, one would need to digest Google, books, and wait for a mentor/colleague to point you in the right direction. Taking hours if not days.

20+ years with JavaScript. What a great decision. (To be fair.. thanks PHP too)

The desire for a debugger in Zed grew into our most upvoted issue ever. Today, the issue has accumulated 2143 👍s. Nothing but respect to `console.log()` warriors, but engineers clearly wanted to be able to line-by-line step through their software and inspect state.

@github.com would love if the diffs the code was collapsable

If I had a penny every time I had to wipe out node_modules. I'd have a solid bitcoin.

One would think that Google Sheets has an import from JSON in 2025.

Who cares about AGI. I'd kill for automatic git conflict resolution.

TIL

🤦‍♂️

I would love if my IDE highlighted public interfaces. i.e. top level exports and top level return statements inside of functions. Anything that would quickly help me understand the established contracts.

Outlines is a good start, but needs tweaking to improve the signal to noise.

Always fun to threaten the LLMs that you will switch to another model if it doesn't try harder.

My guilty pleasure with prompting LLMs: no need to correct grammar or spelling mistakes.

Mutation testing may just be the best way to slop-proof your code.

Best of both worlds:

- AI helps you ship (vibe code) fast
- Mutation testing helps you check the slop(py) tests that they're actually exercising code

Time to pick up new tools.

I would kill for a TypeScript equivalent of Python's decorators.

"Told you so" -- I'm sure -- a lot of people are saying now.

Just look at function calling using JSON schema (and Pydantic or Zod). The more we can declare the intent and relations the better the AI performs on your behalf.

The best part of this revolution is seeing well known practices become popular all over again.