Daily AI news. I build with it too.
Chennai
Joined September 2010
- Tweets2.7K
- Following1.7K
- Followers384
- Likes2.3K
The fastest way to stop an agent fixing the wrong bug: make it fail first.
Before any fix:
1. write a test that reproduces the bug
2. run it, show it fails for the reported reason
3. fix
4. rerun it plus the suite
No red test, no fix.
Which step does your agent skip?
An AI code reviewer with RAG did worse than one with no added context.
Compare the Market, 79 real merge requests:
- GitLab Orbit graph: ~70% of expected comments
- RAG: ~58%
My take: who changed the code beats what looks similar.
What does your reviewer see?
GitHub Agentic Workflows is GA: agents in Actions, written in Markdown. Day one:
gh extension install github/gh-aw
gh aw add-wizard githubnext/agentics/repo-status
gh aw compile
gh aw run repo-status
gh aw logs repo-status
gh aw forecast
Commit both the .md and the .lock.yml.
A lost race doesn't crash Linux, it only fails the handshake. So an unreleased Claude model kept retrying: 48 min to leak a kernel pointer, 5 more to root on an MPTCP server.
CVE-2026-98432, CVSS 9.8, kernels 4.4+
Fixed: 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56
3 more weeks of RL took Mistral Large 4 from 28% to 41% on Terminal-Bench 4.0. Weights are out.
- ~1T params, ~49B active
- 1M context
- 47% fewer tokens per task than GLM-5.3, per Mistral
My take: top of the open pack, still ~20 points behind Mythos 5.1.
Self-host it?
Google now lets outside apps plug into Search and Gemini through one MCP server. 50+ apps are already live in Gemini.
My read: within a year, your MCP server matters as much as your SEO did. Agents pick tools, not blue links.
Too early, or already true?
Claude Code's 3-5 word status line ate 16.4% of one team's weekly spend. The main chat took 15.7%.
- each active background agent resends its full transcript every ~30s
- Anthropic: ~2% typical, 6% in terminal
How many background agents do you run?
Theo asked if T3 Code should manage skills, MCP servers and global AGENTS.md/CLAUDE.md. 960+ likes, 320+ replies.
My read: within 6 months every coding agent app ships a harness manager. Models swap monthly. Your setup is what you keep.
What gets standardized first?
Someone compiled a working RISC-V CPU into Qwen3.5-4B's weights. No training: a compiler sets every weight.
It runs C and MicroPython inside the forward pass and ran 18M instructions with zero errors.
@grok how can DeltaNet state work as exact RAM here?
nitter.cf/divagr1925/status/2115…
The best AI agent passes 10.8% of real tax jobs. Score, cost per task:
- Claude Sonnet 5.5: 10.8%, $10.71
- Fable 5.1: 9.6%, $15.37
- Opus 5.5: 9.2%, $12.96
- Muse Wonder 1.0: 6.0%, $5.60
- Grok 4.7: 5.0%, $4.57
- GPT-6 Astra: 4.8%, $13.88
Surge DAYJOB: Tax, 95% rubric bar.
A reviewer subagent that inherits your chat shares your blind spots. Matt Pocock caught Codex forking parent context into subagents.
Paste:
"Review with a fresh subagent: no chat history, just the diff and the spec."
Do yours start clean?
nitter.cf/mattpocockuk/status/21…
Just shipped an important skill fix for Codex users.
Codex was biasing most of its subagents to inherit its context window from the parent agent. I.e. forking it.
This has a big negative impact on a lot of my skills, that rely on a clean context window (for code review, implement-spec etc)
So, a fix:
github.com/mattpocock/skills…
Android Studio now runs Claude, Codex and 40+ more coding agents in the IDE.
Rabbit 2:
- any ACP agent plugs in
- agents get builds, emulator, Logcat
- Google's test: UI tasks 77% faster with CLI previews
My take: IDEs are now agent hosts.
Which one?
nitter.cf/AndroidStudio/status/2…
You wanted agent & model flexibility, so Android Studio now supports Bring Your Own Agent via the open Agent Client Protocol.
Install 40+ ACP-compatible agents directly in the IDE and pick your preferred models.
Download Android Studio Rabbit 2 today → goo.gle/4iigVuN
Live: webtk.dev/sites/abyssal-desc…
Stack: Three.js + React Three Fiber, GSAP, Lenis, custom GLSL shaders.
Want the full prompt? DM me "ABYSSAL" and I'll send it over.
Pasting an API key into agent chat leaves it in the transcript. T3 Code now asks out of band:
- paste in a box, not the chat
- saved to a private file
- agent's commands can read it for 24h
My take: every agent needs this.
Where do your keys live?
nitter.cf/theo/status/2115743790…
40%+ of GitHub pull requests now start with an agent. In March it was 17%.
Octoverse 2026:
- 1.3B merged PRs in a year, 3x+
- new repos with an agent config: 8x the commits
- 90% of agent PRs merge within 2 weeks
My take: review is the bottleneck now.
Who reviews yours?
Some ChatGPT users saw a pop-up saying Chat and Work are now one. Tibo: a mistake, "not rolling out just yet."
Brockman said in July the Work tab goes by year end.
My read: it ships before December and the model picker disappears with it.
When do you think it lands?
10 people. 77,000 substantial changes in 30 days, about 108 an hour.
Spawn founder Jacob Sansbury:
- 30x payroll spent on AI tokens
- agents fix every error a user hits
- and verify their own work, no human check
Would you ship agent fixes unreviewed?
nitter.cf/jsnnsa/status/21156233…
