Charles Green

Systems, not guesswork.

[Charles Green @charles_green]

No feedback = no learning.
No learning = your AI today is exactly as good as it was 6 months ago.
The teams pulling ahead decided on day one to capture every thumbs up, accepted edit, retry, and "actually nevermind."

[Charles Green @charles_green]

Every agent interaction has two parts:

What the agent did
Whether it worked

Most teams only store #1.

[Charles Green @charles_green]

Most teams shipping AI aren't building products.
They're building demos that have been left running.
The difference: feedback loops.
If you can't tell me which 10% of your agent's interactions last week were most useful, you don't have a product. You have a demo.

[Charles Green @charles_green]

A security firm scanned 1M AI services on the open internet.

Most shipped with no auth. That’s the default.

Your AI agent is a skeleton key, it inherits every permission of every tool you connect it to.

Trick the AI, you get everything it touches.

What can your agent do if someone tricks it?

[Charles Green @charles_green]

Rob Pike gave us Go Proverbs. We needed our own for AI agents.

Five of the 15 aixgo Proverbs:

» A single binary is better than a thousand dependencies.
» Type safety at compile time beats hope at runtime.
» Channels orchestrate agents; gRPC orchestrates systems.
» If it doesn't compile, it doesn't ship.
» Where prototypes go to die, Go agents ship and thrive.

Production AI deserves production tooling.

Full list → https://t.co/RaepU36coY

[Charles Green @charles_green]

"Don't prototype in Python and hope it ships; ship in Go and know it works."

One of 15 aixgo Proverbs we just published.

Every one is a production lesson paid for in incidents — ours and others.

https://t.co/TWh9wGhwTg

[Charles Green @charles_green]

Most AI agent frameworks demo well and scale badly.

Seconds-long cold starts. Gigabyte containers. Vendor lock-in baked in.

Got tired of it. Built the one I wanted.

https://t.co/w6kwBkLmFD — open-source Go framework for production AI agents.

→ Sub-20MB binaries
→ Sub-100ms cold starts
→ Multi-provider, no lock-in
→ Ships clean to Cloud Run, Lambda, edge

Foundation for a few things I'm shipping next. More soon.

[Charles Green @charles_green]

Python wins research. Go wins production.

That split is solidifying fast — Benthos, GoMLX, Argo, Kubeflow, every serious inference layer I see in 2026 is Go.

But nobody’s built a real production-grade agent framework for it. The Python frameworks don’t port cleanly — concurrency, single-binary deploy, predictable latency are exactly where Python struggles and Go shines.

That’s why I’m building https://t.co/w6kwBkLmFD. Open source. Multi-provider. Built for the layer that actually serves users, not the layer that trains models.

The agent serving problem is a Go problem. Most teams haven’t realized it yet.

[Charles Green @charles_green]

Quiet on here for a few months. Tried being an employee again.

Great company. Good people. Wrong shape of work for me.

I'm at my best seeing an org end-to-end, finding the leverage point, building the system that fixes it. Then moving on.

Going independent again next month. AI agents + https://t.co/w6kwBkLmFD.

Follow along.

[Charles Green @charles_green]

19 years in Tokyo across fintech, banking, and Big 4. The last few of those building production AI agents. Both as a CISO making sure they don't blow things up, and as an engineer shipping them.

I'm taking on 2 advisory engagements from May to put it to work.

Who this is for:
* Teams deploying AI agents in production
* English or Japanese-operating, anywhere
* AI ambitions wanting support to deploy at scale

Happy to talk through what you're building. If that's you, send me a DM.

Systems, not guesswork.

[Charles Green @charles_green]

@TalAdam Nice to see Go staying strong at no. 3.

[Adam Tal 🚀 @TalAdam]

I went through every Hacker News "Who is Hiring?" thread since 2019.

83 months. 109,462 listings.

A few things stood out 🧵

[Charles Green @charles_green]

Switch from OpenAI to Anthropic to Gemini.
No code changes. Just update the model name.
https://t.co/kTW5VtKC4o auto-detects the provider.

[Charles Green @charles_green]

Start on Cloud Run. Scale to Kubernetes. Same binary. Same code.
https://t.co/kTW5VtKC4o doesn't force architecture decisions before you have revenue.

[Charles Green @charles_green]

Python AI frameworks fight the GIL to run agents in parallel.
Go was literally built for this.
Goroutines > async/await gymnastics.

https://t.co/ZbDNC0yggt

[Charles Green @charles_green]

Your API gateway boots in milliseconds.
Your database proxy boots in milliseconds.
Why do your AI agents need 30 seconds?
They don't.

[Charles Green @charles_green]

One binary. Zero dependencies. 8+ LLM providers.
go get https://t.co/lyX0eXO0kE
That's the whole install.

[Charles Green @charles_green]

Python: "I'll catch that error in testing"
Go: "I caught that error at compile time"
Your AI agents deserve the same.

[Charles Green @charles_green]

Your AI agent doesn't need a 1GB container.
https://t.co/kTW5VtKC4o ships as an 8MB binary.
Same agents. 125x lighter.

[Charles Green @charles_green]

Aixgo is MIT licensed. Fully open source.

Get started in 60 seconds:

go get https://t.co/lyX0eXO0kE

📦 GitHub: https://t.co/lyX0eXO0kE
📖 Docs: https://t.co/7YNmPRBtFY

#golang #AI #LLM #OpenSource

[Charles Green @charles_green]

Introducing Aixgo v0.3.3 Beta

The AI Agent Framework That Ships in 8MB.

Build, deploy, and scale AI agents in Go.

No containers. No cold starts. No Python.

Your AI infra just got 125x lighter.

Thread on why this matters for production AI...

[Charles Green @charles_green]

13 orchestration patterns out of the box:

→ Supervisor
→ Parallel (3-4x speedup)
→ RAG
→ Reflection
→ MapReduce
→ Router (25-50% cost savings)
→ And 7 more...

Production patterns. Not toy examples.