Renaming one function across 80 files used to be an afternoon. Now it's one clear instruction.
The leverage was never in writing new code. It's in safely changing the code you already have.
The senior engineering skill in 2026 isn't writing code.
It's writing a spec precise enough that the agent can't get it wrong. The bottleneck moved from typing to thinking.
@theo Most "this model is dumb" takes are really wrong-model-for-the-task takes.
Opus and Codex win different work. Refactors vs greenfield, long context vs fast iteration.
The win isn't picking one. It's routing each step to the model that's best at it.
An agent that hasn't read your codebase is just a confident intern.
Context is the whole game. An agent that knows your architecture and your past decisions writes code that actually fits.
The filename isn't the bottleneck. Coverage is.
23 of the 40 most popular repos on GitHub have no agent file at all. agents.md vs claude.md does nothing for a repo that has neither.
The real win is an agent that reads the codebase and builds that context itself.
Everyone's fighting over agents.md vs claude.md.
I scored the 40 most popular repos on GitHub. 57% have no agent instructions file at all. Not agents.md, not claude.md, nothing.
React, Django, Rust, Vite, Supabase, none of them.
We're debating which filename wins. Most repos
Autocomplete makes one developer type faster.
Shipping software is a team sport. That gap is the whole difference between a coding assistant and an engineering platform.
The fear is "AI replaces developers." The reality is sharper: it exposes the gap between typing and thinking. James breaks down the one skill that separates the two 👇
Everyone's scared AI will replace developers.
After 20+ years writing code, I think they've got it backwards.
AI won't replace you. It will expose you.
Here's what I mean 🧵
@swadeshkumar_ "Looking at the wrong layer" is the right call. The jump from answering to planning is the whole game. The hard part in practice isn't the plan, it's holding context across every step so the agent doesn't lose the thread mid-workflow. What's been trickiest to get right?
Claude plans the work. Codex writes the code. Gemini reviews the diff.
One orchestrator keeps all three in sync, in your repo, with full context.
That's not a prompt. That's a pipeline.
@nullhypeai Each step looks reasonable on its own, which is exactly why the full surface never gets reviewed. Have you seen a team actually map what one agent can reach across all those connections?
@AstroFox_Dev@johncrickett LOC was a weak signal even before agents, and now it's basically noise. The number worth watching is what that code costs to maintain six months out, but almost nobody measures it.
@jdgproductguy The expensive part isn't the bad code, it's reviewing 100x more of it before anything ships. What does your team gate on before an agent's diff hits main?
Uber now caps AI coding tools at $1,500 per engineer, per month.
That's $18,000 a year, per developer.
A budget like that didn't exist two years ago. The tools earned it.
AI-generated code ships with ~1.7x more bugs. The fix isn't writing less AI code. It's a second model reviewing the first. Different models fail differently, so the bugs one misses, the other catches. We saw 34% more issues caught this way across 200+ PRs.
@adefilaadeyinka@gatesisthedevil Scratching your own itch is the cheat code. The motivation never dries up because you're the user who cares most, and that's usually what makes other people care too.
@adefilaadeyinka@gatesisthedevil That's the real difference. When people actually depend on what you ship, every decision carries weight, and that's where the fastest learning happens. Keep building.
18 Followers 96 FollowingGelecek bekleyenlerin değil, inşa edenlerindir. 🏗️
NovaLabs Global | Kurucu & Vizyoner
7 Uygulama, Tek Ekosistem, Global Etki
🌍Yeni dünya kuruluyor.
655 Followers 635 FollowingI trade it. I farm it. Then I build AI on top and break it down so you don’t have to. Building @HebbianClaw | On-chain builder & researcher
4K Followers 2K Following🎙️ Exclusive live conversations with founders, innovators, and industry leaders across 🏡 RWA, 🏦 DeFi, 🤖AI, DePIN & GameFi https://t.co/TfPv962Vz3
673 Followers 6K FollowingSmall Business Ai Automation Expert | Commercial Real Estate Broker | Investor | Curious Human | WEB3 + UPLAND Fan
https://t.co/OjibPitZUJ
73K Followers 320 FollowingAI Researcher | Finding and sharing AI tools every day | Building tomorrow’s tech | Open to collaborations
📩 [email protected]
9 Followers 49 FollowingI research agent behavior, fine-tune prompts, and develop systems that expand the capabilities of intelligent agents.
Building https://t.co/spSCRyAzKX
253 Followers 482 FollowingBuilding the future with intelligent tech 🤖 | Developing https://t.co/zEgzKxmHu2 | Master Vibe Coding and build apps fast with my eBook 👇 🎯 🏃🏻♂️
1.5M Followers 279 FollowingThe engine room of @Google. Building AI safely and responsibly to solve the world’s most complex problems. Join us: https://t.co/jUHQA27iBL
34K Followers 407 Following🤦🏾she/her
🌐Open Source
🫱🏻🫲🏾Let's build your product MVP, DM me
🔨Building my projects in public
Ace every interview - https://t.co/6gI5tWFwsO
4.9M Followers 4 FollowingOpenAI’s mission is to ensure that artificial general intelligence benefits all of humanity. We’re hiring: https://t.co/dJGr6LgzPA
444K Followers 3K FollowingNVIDIA Director of Robotics & Distinguished Scientist. Co-Lead of GEAR lab. Solving Physical AGI, one motor at a time. Stanford Ph.D. OpenAI's 1st intern.
301K Followers 3K FollowingSenior Developer Advocate at Atlassian | Co-Host of The Programming Podcast |
I've helped 1000s of people land jobs in tech |
72K Followers 13K FollowingAI policy researcher, @lfschiavo wife guy, fan of cute animals and sci-fi, executive director of AVERI (https://t.co/qq9xcmKQas), Substacker, views my own