Fernando Taricano

Shipping production agentic AI

[Fernando Taricano @TaricanoF]

a lot of people still use AI as a side tool.

what happens if you try doing everything through it?

run your daily workflows inside Claude Code or Codex. every time something works, turn it into a skill. every recurring workflow becomes a scheduled task.

then add a second brain, scoped across the different domains of your life: people, companies, projects, processes, decisions, conversations.

over time, you’re not just automating tasks. you’re building a persistent context layer that knows how your life and work actually operate.

the more you use it, the more useful it gets.

[Fernando Taricano @TaricanoF]

@BarnCodes Hey, let’s connect!

[Fernando Taricano @TaricanoF]

@nix7251 Let’s connect !

[Fernando Taricano @TaricanoF]

Today, I’m building a tiny MCP so my agent can plug into any harness and handle payments safely. Scoped tokens, tight limits, no raw card data.
Would you give your AI agent access to your credit card?

[Fernando Taricano @TaricanoF]

@bcherny For the MCP servers I build, the skill needs to explain failure behavior as clearly as the happy path: partial results, pagination, and which writes are safe to retry. I'd love plugin review to test whether those instructions match the actual tool behavior.

[Fernando Taricano @TaricanoF]

@hwchase17 For resumable workflows, I'd persist the routing decision with its input and model version. A checkpoint replay shouldn't silently make a different choice because the router changed while the run was paused. How do you handle that boundary?

[Fernando Taricano @TaricanoF]

@omarsar0 The boundary I'd keep outside the self-modifying code is authority. Recursive invoke calls should inherit a narrower tool scope, never expand it. Can JAZ keep approval checks outside the code the agent can rewrite?

[Fernando Taricano @TaricanoF]

@noahzweben A regression test I'd add: user A retrieves a document, loses access, then asks a follow-up in the same thread. Can cached context still surface that document? Revocation needs to cover retained context as well as the next connector call.

[Fernando Taricano @TaricanoF]

@danyulol One failure case I design around: a write succeeds, but the tool times out before returning. Retrying can duplicate the action. I want the harness to read back the resulting state before deciding whether another write is safe.

[Fernando Taricano @TaricanoF]

@sarthakaag @X Let’s connect

[Fernando Taricano @TaricanoF]

@elgermerlo Let’s connect

[Fernando Taricano @TaricanoF]

@AliShazil_ Let’s connect

[Fernando Taricano @TaricanoF]

@hrshsn_ Let’s connect

[Fernando Taricano @TaricanoF]

been working on making context rotation automatic for my agents.

once a session crosses 300k tokens on my opus 5.5, it spins up a fresh one, creates a detailed handoff, waits for the new session to accept it, then kills the old one.

so far it feels much cleaner than manually managing long-running sessions.

[Fernando Taricano @TaricanoF]

Which platforms do you guys use for agent orchestration, and why?

[Fernando Taricano @TaricanoF]

@moneyzqux @X Let’s connect

[Fernando Taricano @TaricanoF]

@hasun_bandara @X Let’s connect 🤝

[Fernando Taricano @TaricanoF]

@benz__002 Let’s connect

[Fernando Taricano @TaricanoF]

@VPanev27435 Pretty cool. Let’s connect 🤝

[Fernando Taricano @TaricanoF]

@OmNawale45831 @X Let’s connect 🤝

[Fernando Taricano @TaricanoF]

@whoisdivyam Let’s connect 🤝