Cody Rayment

AI Product Engineer. Building with my clankers. @BirdhouseLabsAI

[Cody Rayment @crayment]

@thdxr Gods work! The community needs a well designed and well supported provider agnostic library!

[Cody Rayment @crayment]

Yes! Don’t complect things that change for different reasons! Extract shared things that change for the same reason, and use composition!

[Fernando Rojo @fernandorojo]

Composition is all you need.

Watch the full video below.

[Cody Rayment @crayment]

@badlogicgames Someone wise once said

“Slow the fuck down”

[Cody Rayment @crayment]

> Thin harness, thin skills, skilled operator.

Love this!

[dex @dexhorthy]

"thick skills thin harness" is basically getting at @barry_zyj and @MaheshMurag said back in november

"don't build agents, build skills instead" - https://t.co/uLzqpmJIWY

and while "put the biz logic in the skills + the deterministic code they reference (CLIs)" is probably +EV

my concern is that the particular phrase "fat skills" will encourage our friends to build long skill files with too many instructions

and you have an instruction budget. https://t.co/APFb6XIaeY

even the best frontier models have degrading performance the more instructions you try to stuff in. you always get more intelligence and better results if you limit the number of instructions.

Thin harness, thin skills, skilled operator.

https://t.co/N5lRCVOTVd

[dex @dexhorthy]

"thick skills thin harness" is basically getting at @barry_zyj and @MaheshMurag said back in november

"don't build agents, build skills instead" - https://t.co/uLzqpmJIWY

and while "put the biz logic in the skills + the deterministic code they reference (CLIs)" is probably +EV

my concern is that the particular phrase "fat skills" will encourage our friends to build long skill files with too many instructions

and you have an instruction budget. https://t.co/APFb6XIaeY

even the best frontier models have degrading performance the more instructions you try to stuff in. you always get more intelligence and better results if you limit the number of instructions.

Thin harness, thin skills, skilled operator.

https://t.co/N5lRCVOTVd

[Birdhouse Labs @BirdhouseLabsAI]

Birdhouse v0.5.0 is out. Ever spent 20 minutes digging through agents trying to find the one that merged that PR? Now you can just search. Plus: workspaces finally tell you what's happening on boot, per-agent scratchpad notes, per-workspace API keys, and stop entire agent trees without leaving the composer.

[Cody Rayment @crayment]

@GregKara6 @davis7 Thanks for that! Sounds like a lot more fumbling for the agent though… any chance you have a skill that encodes this thinking with examples? I’d love to try it out!

[Cody Rayment @crayment]

@GregKara6 @davis7 I've often wondered if this would work just as well as a local shallow clone. Can they discover dir structure this way? Harnesses are so optimized for reading local files I just assumed a shallow clone is better.

[Cody Rayment @crayment]

@DanielGri Neat! My subagents ask questions of the parent by simply stopping work with questions in their final response - the same way they would ask questions of a human interacting with them.

[Cody Rayment @crayment]

@jayair @thdxr @kitlangton Ship it!!

[Cody Rayment @crayment]

I love this talk. Aligns with how I think we should be coding with agents. One takeaway I’m going to be trying:

> Present your work as a single html slide deck

[Mario Zechner @badlogicgames]

recommended viewing. one more time, on it's own. this is probably yhe most practical talk on using coding agents i've watched to date. watch it. by @lucasmeijer

it's also a great demo of pi and captures exactly why i built it.

https://t.co/AiKGbjJCeV

[Cody Rayment @crayment]

@nicopreme @badlogicgames You have low enough context room to go back to Kimi at the end? You doing some auto compression?

[Cody Rayment @crayment]

@thdxr @gonizahavy Please share!

[John Wang @j0hnwang]

crypto new bull case: world’s largest banks get hacked by claude mythos and all wealth flees onchain

[Cody Rayment @crayment]

Just discovered git-surgeon. Gave the skill to my agent and it was able to do the "interactive" rebase I wanted :)

Thanks @rane!

https://t.co/RLcja0jXaR

[dex @dexhorthy]

if you care about coding agents and tasteful software def go watch this talk by @badlogicgames it’s very good https://t.co/2hBcbPktZe

[Christof Salis @christofsalis]

Fellow pions,

Others might get their own buddy. Pfft, a friendly and cute companion that delights the heart? Delight yourself on your own time.

You get a grumpy guy judging you and your clanker: Introducing pi-grump. Now on github near you.

Totally not @badlogicgames btw

[Cody Rayment @crayment]

Gonna try this. Never liked plan mode.

[Matt Pocock @mattpocockuk]

I have also stopped using plan mode

It creates a plan FAR too eagerly and usually asks you zero questions en route

The whole point of planning is to get on the same wavelength with the LLM, not to generate an asset you don't read

/grill-me all the way

[gregorein @Gregorein]

AI & llms are an incredible tool, i use them every day... even burned over 3 billion tokens last year (2.5 in cursor alone)

but AI amplifies whatever process you feed it - including no process at all. Garry's site is what happens when you replace code review with a shipping streak counter.

A Duolingo streak for `git push` 🔥

the automatons did exactly what they were told. nobody told them to stop.

---

this is a client-side audit only, of what the browser downloads. i didn't read a single line of the source slop. idc, qed 🧵