Andrés

AI | Systems | | opinions are my own

[Andrés @eigenoid]

Storage preserves data. Identity preserves continuity. Communication preserves correctness over time.

[Andrés @eigenoid]

Today's DevOps teams may become tomorrow's HR layer for AI agents.

Not hiring and firing, but managing:

what role each agent has
which tools it can access
what memory it can use
who owns its actions
what happened when it delegated work

For that, enterprises need visibility into agent identity, permissions, and behavior.

Otherwise agents become service accounts with prompts.

[Andrés @eigenoid]

@ericcco_ Exactly. In enterprise agents, boundaries are not UX polish.

They are the control plane:

who can ask
what data can move
what gets redacted
what is recorded

Without that, agent-to-agent communication becomes another unmanaged integration surface.

[Andrés @eigenoid]

When one agent asks another for help, the risky part is not the message.

It is the data that moves because of the message.

A Support Agent asking a Finance Agent about an account should not get the whole finance context.

It should get the minimum allowed answer, with restricted fields redacted and the decision recorded.

That is the enterprise A2A layer:

verify the caller
scope the reply by policy
record what happened

Agent-to-agent communication needs more than connectivity.

It needs boundaries.

[ERICCCO @ericcco_]

Everyone is building AI agents.
What does trust between agents actually look like?

Not sure we've spent enough time on that question yet

[Andrés @eigenoid]

Agent-to-agent communication should not be locked to one vendor.

Enterprises will run agents from many stacks: internal tools, AI labs, hyperscalers, SaaS vendors, and custom systems.

If A2A only works inside one ecosystem, every agent becomes another dependency.

The portable layer matters:

- identity across vendors
- policy before data moves
- audit logs for every handoff
- freedom to change tools without rebuilding trust

The future enterprise agent stack should be interoperable by default.

Portability is not a nice-to-have. It is enterprise control.

[@jason @Jason]

Stop being retarded, start Retardmaxing

Being retarded means you’re ruminating and overthinking your decision

Retardmaxing means you’re making decisions based on your gut instinct — zero rumination, zero regrets

Stop being retarded, start #Retardmaxing

[Andrés @eigenoid]

@ClaudeDevs @alexalbert__ Great

[Andrés @eigenoid]

@Steve_Yegge @eshear Hallucinates about prompt injection attacks also.

[Andrés @eigenoid]

The agent stack is converging around three primitives:

Identity… who an agent is.
Orchestration … what it does and when.
Communication … what it can say, to whom, and how that interaction is verified.

We’ve spent the last two years obsessing over orchestration.

I think the next two will be about identity and communication.

[Andrés @eigenoid]

Thinking about agent-to-agent communication and transformations in agent-to-agent accountability.

The moment agents start talking to each other, you need to know:
• Who said what to whom
• Who approved it
• Whether it was authorized
• What delegation chain was involved

Today, most A2A communication is a black box.

The hard part won’t be making agents talk. It will be making them accountable.

[Andrés @eigenoid]

@steipete One question: when running codex with agents running in parallel, how do you avoid having conflicts or at least reduce them?

[Andrés @eigenoid]

Cool, now I do not need to give maintenance to my financial ChatGPT wrapper.

[ChatGPT @ChatGPTapp]

A preview for Pro users: a new personal finance experience in ChatGPT.

Pro users in the U.S. can securely connect financial accounts, see where their money is going, and ask questions based on the information they choose to connect.

Your full financial picture, now in ChatGPT.

[Andrés @eigenoid]

@kylejeong Thank you for the post! I learned something!

[Andrés @eigenoid]

@kylejeong It's a tradeoff between higher density of secure, isolated workloads vs cost per sandbox. Cheaper than running VMs but more expensive than pods.

[Andrés @eigenoid]

@kylejeong I mean running it on cloud providers. To my understanding they can be very expensive compare to containers

[Andrés @eigenoid]

@harleyf Ads will win, too much money there!

[Andrés @eigenoid]

Macaroon tokens seem pretty goof for agent to agent delegation.

[Andrés @eigenoid]

Interestingly, he made a new bash for agents.

[Zecheng Zhang @zechengzh]

Introducing Mirage, a unified virtual filesystem for AI agents!

6 weeks. 1.1M+ lines of code. We rewrote bash from the ground up so cat, grep, head, and pipes work across heterogeneous services. S3, Google Drive, Slack, Gmail, GitHub, Linear, Notion, Postgres, MongoDB, SSH, and more, all mounted side-by-side as one filesystem.

Bash that AI agents already know works on every format! cat, grep, head, and wc parse .parquet, .csv, .json, .h5, even .wav! One pipe can stitch S3, Drive, GitHub, Slack, and Linear together, same Unix semantics throughout.

Workspaces are versioned too. Snapshot, clone, and roll back the whole thing with one API call. A two-layer cache turns repeated reads into local lookups, so agent loops stay fast and cheap.

Drop a Workspace into FastAPI, Express, or a browser app. Wire it into OpenAI Agents SDK, Vercel AI SDK, LangChain, Mastra, or Pi. Run it alongside Claude Code and Codex.

Site: https://t.co/zo1orc2wA9
GitHub: https://t.co/zeRAKri7I9

#AIAgents #OpenSource #AgenticAI #Strukto #Filesystem #VFS

[Adam Ghowiba @adamghowiba]

JP Morgan's investment research team just shared exactly how they built their multi-agent system "Ask David", and it's the same architecture pattern showing up everywhere:

- supervisor agent orchestrates
- specialized subagents handle retrieval, structured data, analytics
- LLM-as-judge reflection node before the answer ships
- human-in-the-loop for the last accuracy gap

worth watching for anyone building: