Jarrod Watts

Jarrod Watts

23,052 Twitter followers

Follow

Posts

1M context window, memory updates, and the Australia buff (double off-peak usage) have got me hooked on Claude Code again

Today feels like a good day to share this again twitter.com/jarrodwatts/status...

Plugins are coming to Codex (soon). They're packaged bundles of skills, MCP servers and hooks (also coming soon) that extend what your agents can do. Including an "openai-curated" collection with 200+ integrations out of the box - Slack, Gmail, Figma, Stripe, Linear and more.

code review tools are pretty annoying rn because they tend to always find *something* wrong. they don't ever seem to reach the point of "lgtm". then you get stuck in this weird cycle of: agent writes code → agent finds bugs → agent writes code → agent finds bugs → repeat

I’m fully CLI pilled now. They’re now the most important surface of a developer tool. Agents are VERY good at understanding CLIs, much better than humans are. For example, I’m now an avid user of Railway, solely because my agents are so good at using their CLI to manage infra.

Codex has a hidden superpower - multi-agents. I made a quick demo showing how it works: · Enabling the experimental feature · Spawning multiple agents in parallel · Talking to sub-agents directly

The OpenAI Engineering blog is probably the single best source of information out there for AI currently. I strongly recommend spending 10 minutes less doomscrolling 𝕏 today and 10 minutes more reading at least one of their posts. Insanely valuable information in there.

Loading..