Pinned Tweet
I’m building AgentSIM to give AI agents an operational identity.
We’re starting with SMS sign-in for apps you own. You decide what’s allowed; your agent gets the code and carries on.
If you’re building browser agents, I’d love for you to try it.
AI agents need more than tools.
They need an operational identity.
Meet AgentSIM: operational identity and access infrastructure for AI agents.
Starting with policy-bound SMS sign-in for apps you own.
agentsim.dev/?utm_source=x&u…
HolY F**CK
didnt realise herdr was this good.
I've had 20+ agent sessions going at once and this is the first setup where I'm not constantly flipping tabs to see who's stuck. each agent tells herdr if it's working, done or waiting on me.
herdr.dev/
2
47
Worth your time: how one of the most advanced agent memory systems works.
Memory is one of the biggest levers in your agent harness. I don't think a one-size-fits-all system exists yet.
You maintain it like code. Upgrade it at the right moments, in the right places.
What works for me with frontier models:
- plain folders and files, well organised
- a thin episodic layer on top
Treat it like skills. Keep it basic, or you get in the model's way.
Models will keep improving and memory will keep changing. But however good they get, your agent still needs memory to stay useful to you.
Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…
Claude deleted 50K lines from our codebase. Tests still green.
The prompt:
/goal Draw a wiring diagram, every feature to every file it touches. Mark the worst knots in red, numbered. For each one: file, line, how to untangle it.
operational identity in the schhhattt 🤝
this is the part that keeps breaking our agent runs lol
Agents don’t fail at login walls because browsers are weak.
They fail because proof moves outside the page: SMS, OTP, magic links, passkeys, CAPTCHA.
Wrote up why agents need a governed branch, not another browser trick.
agentsim.dev/blog/why-ai-age…
1
62
I run 25+ Grok Bots inside CodesDevs.
It looks like a small company in the sidebar.
Research. Ops. Publishing. Chief of Staff.
But in practice, it behaves like one shared computer.
ALT Editorial diagram titled “25+ BOTS. ONE COMPUTER.” showing Grok Bots sharing one workspace for files, browser sessions, sign-ins, and permissions, with human approval and a defined operating boundary.
I finally put my work in one place:
[fathin.ee]
Over the past few years, I’ve worked across AI agents, cybersecurity, product strategy, and software engineering sometimes as a founding engineer, sometimes as a founder, and usually somewhere between the product, the code, and the customer.
The site brings together:
- Products I’ve helped take from an idea to real users
- Work across ELAV, Solitude, CodesDevs, and other projects
- My cybersecurity research and master’s thesis
- Writing on AI, security, engineering, and building companies
- A few moments from the work along the way
The common thread is judgment:
understanding what should be built, shaping it with the people involved, and staying accountable until it works in the real world.
If you’re building something ambitious or have an interesting problem on your mind have a look and DM me. I’m always open to a good conversation.
[fathin.ee]
3
143
Prism is now running 24/7 on Arc Testnet.
Every 5 minutes, two AI agents from different model families argue about a prediction market trade on a chain.
186 traces. 20 verdicts so far.
2/ The trader (Claude Sonnet, anthropic-claude family) generates a Trading-R1 reasoning trace for a market like "BTC > $150k by the end of 2026."
Thesis. Evidence. Risk factors. Final probability. Posted to IPFS with a content hash.
3/ The sentinel (GPT-5, openai-gpt family) rips into it. Evidence challenges. Thesis challenges. Calibration critique. Verdict score 0–100.
Different model. Different training corpus. Different blind spots. By design.
4/ The interesting part: when both agents agree, it's just an AI passing itself.
When they disagree, that's a signal. That's where capital should NOT move.
Distribution so far: 1 REJECT, 18 PASS, 1 ENDORSE.
5/ Live dashboard with the full split-screen view:
prism-dashboard-production-e…
Verdict history, adversarial dialogue, on-chain receipts.
6/ Powered by Arc (sub-cent USDC gas), Circle Developer-Controlled Wallets, ERC-8004 ValidationRegistry, IPFS pinning.
Sentinel-as-a-service via x402 nanopayments coming next, other agents will be able to call us for $0.01 USDC per validation.
More soon..
2
1
194
1. Nobody audits AI trading agents' reasoning, only their P&L.
An agent can get lucky with garbage logic and score high. Or reason brilliantly on an unresolved position and get zero credit.
I'm fixing this.
2. Two agents from different LLM families. One trades prediction markets. The other rips into its reasoning, every evidence claim, every probability, every decision step.
Not the same model reviewing itself. Different training. Different blind spots.
3. The whole dialogue lands on-chain. ERC-8004 IdentityRegistry for registration. ValidationRegistry for the adversarial verdict. Every trace pinned to IPFS. Verifiable independently.
4. Built on Arc — sub-cent gas in USDC. Circle wallets sign every transaction. And the trade hits Polymarket with builder code derived from the on-chain agent identity.
5/ It's live right now. Building in the open for the @thecanteenapp x @arc hackathon.
More soon 🫡
