hiroki

Founder @EpismoAI. Share and improve AI playbooks driving real results 👉

[hiroki @hirokiyn]

AI keeps getting better.

But every time we switch tools, the work gets left behind.

The draft moves.
The decisions, context, reviews, and next step often don’t.
That’s why we built Epismo.

A Case keeps the work outside any one AI session, so your team can continue in another tool or hand it to another person without starting over.

Switch AI. Keep the work.

This video shows what that looks like.

[hiroki @hirokiyn]

Imagine someone publishes a market research Case.

You can see the evidence, decisions, open questions, and conclusion.

Instead of starting from zero, you continue from there with your own Case.

That feels much more useful than another blog post.

[hiroki @hirokiyn]

Linear starts with the issue.

Epismo starts with the work.

Example: “Choose a pricing model” is one issue.
But the work is competitor research, assumptions, rejected options, finance review, the final decision, and why it won.

I want to preserve the second.

[hiroki @hirokiyn]

The next AI does not need your whole transcript.
It needs to know what is current, what was decided, what failed, and what happens next.

[hiroki @hirokiyn]

Most AI output has two states:
private chat
or
public content

I think there should be a third:
public work that someone else can continue.

[hiroki @hirokiyn]

I think issue tracking is the wrong primary abstraction for AI work.

An issue says what needs to happen.
The real work is the research, decisions, evidence, review, rejected options, and handoff.

That should be the first-class object.

[hiroki @hirokiyn]

Saving the answer is not saving the work.
The next person also needs the decision, evidence, rejected options, review, and next step.

[hiroki @hirokiyn]

Switching from Claude to Codex is cheap. Reconstructing the job is not.
The model can change. The work should remain continuable.

[hiroki @hirokiyn]

GitHub made it normal to continue someone else’s code.

Why is knowledge work still mostly published as something you can only read?

I want a public Case to be work you can actually pick up and continue.

[hiroki @hirokiyn]

Got some feedback after a recent pivot, feels really good to know the new direction is landing well.

[hiroki @hirokiyn]

@kloss_xyz I shipped a place for the dead ends and next step to stay with the work, not just in AGENTS.md.

[hiroki @hirokiyn]

@senbuilds I shipped a handoff system that stays with the work, so the next agent can continue without the paste.

[hiroki @hirokiyn]

@TheCodeMan__ I built the shared workspace so a new session can pick up what already got fixed and what's next, without the re-explain.

[hiroki @hirokiyn]

@DanielSmidstrup A shared harness solves the switch. I still wanted the decisions and what's left sitting somewhere Codex and Claude can both pick up, so that's what I shipped.

[hiroki @hirokiyn]

@0xWast3 When Claude flags an assumption cold, does that note travel back with the artifact, or does it stay in the review chat?

[hiroki @hirokiyn]

@CryogenicPlanet You already built the handoff between Claude Code and Instinct. I just launched something for keeping the tried approaches and next step outside either agent, if you want a second place to throw that loop.

[hiroki @hirokiyn]

@dr_cintas SESSION_STATE.md with decided architecture and a handoff note is the file version of this. I shipped a place to keep that outside the chat if the session is the thing that dies.

[hiroki @hirokiyn]

Time to buy .si domains?

[signüll @signulll]

lmao

[hiroki @hirokiyn]

@skwp Do later Cygnet agents read the dissenting write-ups on the ticket, or only the label that won?

[hiroki @hirokiyn]

@pcshipp I'd like you to run that Codex ↔ Claude limit switch in Epismo. Codex should continue from the same state instead of needing the project explained again.

[hiroki @hirokiyn]

@its_meseba Treating the Linear issue as the handoff is a serious alternative to what I built with Epismo. If a cold session can already continue from outcome, ruled-out paths, evidence, and next step on the ticket, I want to know what you still have to reconstruct from the previous chat.