Two Anthropic seniors just made Karpathy's loop 1000x better with "Graph Engineering" - dropped 11-page PDF
the shift: the agentic systems got 1000x better the moment you wired agents into a graph
here's the playbook in 6 steps:
step 1 โ build one loop: generate, critique, revise - one self-review cycle beats a smarter model with none
step 2 โ add tools: search, code execution, database - thinking without tools is hallucinating
step 3 โ go parallel: spin up agents in separate worktrees - same repo, different branches, no conflicts
step 4 โ add a graph: agents write findings as typed nodes and edges - not transcripts - every claim keeps its source
step 5 โ ground your evaluator: it checks claims against graph edges, not vibes - "Triple not found" beats "seems off"
step 6 โ the graph survives every session - your agents stop rebuilding context from scratch
the result: Karpathy ran 1 agent in 1 direction - this system runs 1,000 with shared memory - same model, it's the architecture
read this 11-page PDF and paste it into your Claude - you won't regret it
bookmark - then read the article on building graphs from scratch โ
How to graph-max with Codex and 5.6 Sol:
1. Draw a graph (literally in any tool, even on paper)
2. Send it to Codex and say "write a code mode script that implements this workflow, run it with <your inputs>"
There's no step 3, it just works.
Seriously it's that simple.
Built a repo viewer to allow you use ChatGPT or any other LLM with a search tool view any public GitHub codebase so you can ask questions or reference it without setting up GitHub integration. Basically GitHub pages for LLMs. Y'all should try it out. ๐
https://t.co/eHNzUImkY8
Apparently, the best way to gain traction in the tech space is to build something crazy, post it online, gain engagements. One of the following happens:
- Someone big hires you.
- Someone big buys your work.
- Someone big backs your work.
Whichever way, you win.
This my Antigravity be abusing subagents. It's like it doesn't know it's burning tokens ๐ฅฒ It uses it pretty effectively though, especially when I use Claude.
I came across an article I wrote last year and was impressed with myself.
If you're interested in learning how LoRa(Long Range) works, I wrote a comprehensive article on it.
Embedded System and IoT folks will find it useful๐๐พ
https://t.co/gSRZvGVnTp
@chidiebere_x@pamilhereen Dude! You're saying you only care about current paying customers and not potential ones? How's your company supposed to grow like that? I've never used @prymhosting before but it'll to be difficult to do so now. Same for many others. Not that you care, just pointing that out.