PiG retweeted
llama.cpp v1.0.0 is now available
huggingface.co/blog/ggml-org…
PiG retweeted
Happy Birthday Golang,
Go pushed to me to learn distributed systems and Infra,
i don't like solving leetcode questions but i have solved some of them in go, feels way better than cpp to me
over the past 3 months i have also built a repo of all distributed systems patterns written in plain go, entire cncf landscape heavily relies on go because of it's simplicity
my first project which i am proud of is connection pool written in go, connection pool can be used to optimise network latency by picking up already warm connections and avoiding tcp/tls handshake + payload ramp every time
thanks Go, for everything
“Go Turns 17” by Austin Clements, for the Go team — go.dev/blog/17years
#golang
mcp server rn if anyone wants to fuck with it, it’s not bad, just not quite the perfect interface bc of what fl exposes
github.com/zekdevs/fl-mcp
Replying to @PiGCodingAgent @ariaghora
Keep up the work PiG I'm rooting for you 🫵
OPA is surprisingly good and effective at adding another layer of depth to automatic CI checks for your agents. not a lot of people are using it to actually enforce code quality things, but with change detection + AST rules you can get a super fine grained level of depth
PiG retweeted
In this moment, I am Autolithic
autolith.rocks
PiG retweeted
Hi friends! Meet octet.
I rebuilt @pidotdev coding agent in Rust, extensible in any programming language.
Built for speed, with Pi extension compatibility and optional extensions for codemode, subagents, MCP, and more.
free+open source
made with ❤️ in Toronto, ON, 🇨🇦
PiG retweeted
Replayed 144 graded subagent runs from my own sessions. Forked vs. briefed fresh subagent: 100% pass rate on both, so no quality gap here. Forking only won on cost while the main session was under 100k tokens. Past that, ~1.3x, even at 90–99% cache hits.
Subagents are the biggest cause of AI failures
In most harnesses:
1. Main agent writes a summary of what it knows
2. Subagent starts from that summary only, and makes assumptions
3. Subagent works and sends back a new summary of what it did
Same flaw as compaction: all summaries are lossy. This is how your assistant makes emails you never wanted, or purchases what you told it just to get the price of
So I tried something new: checkpoint agents
The initial agent has complete context of what needs to be done. And that input is already cached
Fork the session where it is. Send the fork ahead to do the task and it will do exactly what the main agent would have done; their context is identical
What is even better is when the fork finishes, the cache of the main agent is kept warm by the subagent (they share the same prefix)
I measured it on a 137k-token session with Opus 5.5:
→ Main agent does it on its own: 67 s, $0.38, context +8-17k
→ Checkpoint: 117 s, $0.68, context +2k
→ Classic subagent: 123 s, $0.57, context +4-5k
But here is what I found: avoiding large input is even more important than keeping cache. And most subagent tasks were done better either in series with retrieval, or split up to be parallel over fresh sessions from the start
I do not think there is a place for subagent delegation at all. There should either be a super-agent that orchestrates and does no work, or a long running agent that can retrieve from previous sessions, not a worker that completes tasks partially and delegates the rest to other workers
Don't design harnesses based on our assumptions of human behavior
PiG retweeted
We are looking for teammates who operate with intensity, integrity, and have a yearning to craft software and open protocols that strengthen human agency with us
earendil.com/join/
PiG retweeted
Tern Surface Protocol is amazing.
@can1357 You really did it.
The Obsidian of Terminals & TUIs :)
👀
Replying to @ThePrimeagen
🫡 was gonna be Rust but it really is not there yet iteration speed wise. Go is excellent for this purpose in our evals
PiG retweeted
We're prototyping native language support in Next.js, starting with Route Handlers and Proxy in Go.
This gives you more options for CPU intensive workloads on the server, alongside your React app.
PiG retweeted
people of pi (durable). i did a round of performance optimizations in pi-durable.
quick primer. every step of an agent run is saved to a db before the agent moves on. 1 save = 1 commit. thats what makes it durable, crash mid run, resume from the last commit. ~12 commits per run































