Rishit Singh retweeted
I did what Microsoft didn't in 15 years... Bring back MSN messenger except you are not talking to your friends since you don't have any anymore. you are talking to your agents.
Rishit Singh retweeted
So i studied software engineering just to become a condom between claude and prod 🥀
Rishit Singh retweeted
Harnesses often get dismissed as just scaffolding, just prompt engineering, and not real research. But that couldn't be farther from the truth. The same model weights that score 30% on ARC-AGI score 95% with a better harness.
So we gathered a group of researchers and founders working at the frontier to do a deep dive into the state of harnesses.
We cover how we got to this point, the case for making your harness as expressive as possible, and what YC learned building an agent for every employee in the company.
00:00 - @FrancoisChauba1: Why harnesses matter
04:27 - Building an auto-researcher by accident
07:13 - A five minute history of harnesses
13:56 - Self-improving harnesses
18:35 - @sethkarten: Prime Agent, a self-improving RLM harness
21:50 - Context as an L1, L2, L3 cache
24:51 - From Turing machine to von Neumann computer
28:33 - Messaging between agents
30:04 - ARC-AGI results
33:09 - Emulator Bench and GPU kernels
37:30 - @JonSaadFalcon: OpenJarvis, personal AI on personal devices
38:26 - How far behind are local models
39:21 - The five primitives of a personal AI stack
42:47 - Letting cloud models optimize your local stack
43:53 - 800x cheaper than the cloud
45:58 - @josh__france and @jbellregan: QM, YC's agent harness for work
47:29 - A history of YC's internal agents
49:24 - OpenClaw and a fleet of 50 agents
51:04 - Pulling the brain out of the sandbox
54:43 - Letting the agent choose its own sandbox and model
57:16 - The grind tool: budgets on goals
58:50 - Agents don't understand social context
Rishit Singh retweeted
read code. no matter what. accountability is still yours. optimizing for speed while risking correctness is stupidity.
Rishit Singh retweeted
“The downside of not having a clean codebase is worse than ever.”
just watched an interview from @thdxr explaining why OpenCode’s codebase is now cleaner than it’s ever been.
an agent’s performance is a function of its intelligence, your instruction, and its context (the codebase).
that’s why it’s more important than ever to maintain the codebase regularly, establish patterns, and go back to fix some of the old deprecated ways.
I think the reason models feel worse a while after launch is that a model's performance is a function of its raw intelligence, multiplied by the codebase's quality. yet, every time you use AI, your codebase gets a bit worse. sometimes it is invisible, but confusion accumulates, even when behavior remains intact. unused fields, inconsistent variable names, stale comments. the more the AI edits, the more slop-fried your codebase becomes, like a JPEG that is repeatedly compressed. this, in turn, causes more junk to automatically get in the context, bringing you closer to ugly latent space neighborhoods, making it produce worse outputs, til you reach an equilibrium where no more progress is possible. then a new model launches, and its increased slop-resistance counterweights the accumulated junk, and things suddenly start moving forward again. until your repo crosses that model's slop threshold again, and it stops working, again. I think this process is an inherent part of how these things work and GPT 10 or Fable 10 will still feel moronic after 1 month of launch. the only difference is you'd be maintaining a 100k LOC project, but it will still fail to add a new feature in ways that make you facepalm
Rishit Singh retweeted
young and impressionable engineering students are attracted to the startup space and i think often misled by the supposed glory
all they wanted to do was prove themselves / have a good career / finally make their parents proud. but the sf competition and other unfortunate practices in the sf bubble are lead to more credential chasing, capital chasing, and a surprising uniformity
many of the sf bubble activities i also fell into as a 20 year old on leave from university. taking vc funding early on, immersing myself in the sf community, working 997 and trying to find a place for myself.
but please, if you are young and ambitious in the startup space, stay grounded, don't take bad deals, work hard, and stay a person beyond an empty tech bro.
Rishit Singh retweeted
A billion people produce the most valuable dataset in the world every day – and delete it every night.
We’re recording it.
Introducing @AttentionInc
