Louis
11.6K posts

Louis 리트윗함

have started doing some switch-hitting between claude and codex ever since we rolled our own alternative harness in march. Opus 4.7 now available in humanlayer ofc. Recent vibes:
- opus 4.5 more reliable than 4.6 or 4.7 for most things. (fwiw i actually preferred intelligence of 4.1 to 4.5 but 4.5 is so much faster that its worth the tradeoff)
- 1m context is useful for some cases but i still try to stay under 100k tokens for serious work on our large monorepo by externalizing task progress to MDs
- codex takes more time to review and research code before - this means it needs less intentional steering to research and planning, it can tackle bigger things with just a simple prompt
- codex takes more time, this makes it slower, which can be annoying if you are used to claude just getting to work
- codex, esp 5.4 is less trigger-happy than claude models, it is more likely to ask "want me to run the tests" or "want me to go make this change" vs. just going to do the thing. This actually makes afk yolo-ing harder. but if you are working off plans its fine.
- claude code cli/sdk continues to add features, flag on random new things, we disable a lot like auto memory and adaptive thinking while working. In general the more built in tools and agents, the less room for your instructions. If you don't own the harness you're at the whims of what the model provider thinks you need to get better results.
- this generally seems to be optimized for new users getting results, at the cost of giving flexibility to power users
- the frontier is jagged and opus 4.7 is good at a few random things that prev models were not as good at (no spoilers! go poke around!)
most people in the GC have been saying these things for weeks. but in case you were waiting to hear from me and @0xblacklight and team, here's what we got
thanks @nayshins @GeoffreyHuntley @SickBots @nisten and many others for contributing to the vibe report

English
Louis 리트윗함
Louis 리트윗함
Louis 리트윗함

Been experimenting with procedural locomotion on #GaussianSplat environments from @theworldlabs .
Built a little multi-legged robot in Unity that raycasts against the splat data to figure out where to place its feet, no meshes or colliders involved... Pretty fun to climb surfaces, walk on walls, and set a weird number of legs on the fly :)
Still rough around the edges but pretty fun to watch it figure things out.
#GaussianSplatting #Unity3D #WorldLabs #ProceduralAnimation
English
Louis 리트윗함

At the Agents Anonymous SF meetup last night we did another 🙋 AI usage survey, here are the est. numbers:
Usage stats:
- 90% Claude Code
- 60% Codex
- 30% Cursor
- 20% OpenCode
- 10% Conductor
- 10% Own agent/Pi
80% have prompted a coding agent from mobile
50% have not handwritten a single line of code this year
99% think they're more productive now vs. pre agentic coding agents
Parallel agent usage:
- 90% 3+
- 70% 4+
- 50% 5+
- 5% 10
Also want to give a ginormous thank you to our incredible speaker lineup:
- @jonas_nelle & @alexirobbins from @cursor_ai
- @southpolesteve from @Cloudflare
- @LewisJEllis from @ycombinator
- @aidandcunniffe from Git AI
- 🦞 @steipete from @openclaw
Hope to see you all at the next one! 🫡
English
Louis 리트윗함

What's missing to make AI agents mainstream? For more perspective on putting agents into production for real-world engineering teams, we spoke with founder @dexhorthy of @humanlayer_dev. hubs.ly/Q047cLd80

English
Louis 리트윗함
Louis 리트윗함
Louis 리트윗함










