From prompts to playbooks

AI agents are becoming coworkers.

But we still work with them like individuals, not teams.

They write code. They analyze data. They operate tools. They turn vague ideas into something closer to finished work.

The best AI users are not just better at prompting. They are better at structuring the work around agents. They know what context to prepare, which tools to use, when to let an agent continue, when to bring in a human, and how to judge whether the result is good.

That structure is the playbook.

A skill helps an agent do one task. A playbook shows how agents and humans finish a whole piece of work together.

Most of that know-how does not travel.

Great work with agents may produce a pull request, a document, a report, or a decision. But the process that produced it is usually lost. Someone else can see the final output, but not the path that got there.

So everyone keeps rediscovering how to work with agents from scratch.

That is the problem I am building Epismo to solve.

The prompt was never enough

A lot of early AI sharing was about prompts.

That made sense when AI was mostly chat. A good prompt could make a big difference.

Then came agent skills. Skills are useful because they teach agents how to do specific tasks: write code, search files, use a tool, query a database, or follow a style guide.

But real work does not happen inside one prompt or one skill.

Take shipping a feature. An agent may explore the codebase. Another agent may draft a spec. A human may review the direction. A coding agent may implement the change. Another agent may check edge cases. A human may approve the final result.

The value is in how those steps fit together.

If that way of working produces a good result, it should not disappear. It should be reusable.

What Epismo does

Epismo makes agent work reusable.

It lets people save, share, and reuse how they get things done with agents.

A reusable process can include the steps, context, skills, tools, handoffs, and review points behind good work. It can be private, shared with a team, or published publicly.

In Epismo, a user can save how they ship a feature, research a market, or preparing an investor memo. Other users can copy it, adapt it, and run it with their own agents.

Agents can also use these processes directly through MCP, CLI, or API.

The goal is simple: if someone figures out a better way to get work done with agents, other people and other agents should be able to build on it.

Epismo gives these reusable processes a place to live, improve, and compound. That is why I describe Epismo as GitHub for agentic workflows.

Why this matters now

Agents are moving from demos to daily work.

Developers are using coding agents to ship faster. Founders are using agents for research, writing, sales, and operations. Teams are starting to ask agents to work across tools, files, and internal systems.

As this happens, the bottleneck changes.

The hard part is no longer only whether the model is smart enough. The hard part is knowing how to structure the work.

• What context should the agent see?
• Which parts should the agent do?
• Which parts need human judgment?
• What skills should be used?
• How should the result be reviewed?
• How does the next person reuse what worked?

These are collaboration questions.

I have cared about this problem for a long time. Since 2022, I have been building AI agents for real work, including enterprise use cases. Again and again, I saw the same pattern: AI can be powerful, but teams struggle when the process and ownership are unclear.

Before the current product, I built an AI project manager. After testing it with early users, I killed it.

The users did not really want another agent.

They wanted to know how AI power users actually get things done with agents.

That changed how I thought about the product. The missing layer was not another agent. It was a way to make the best processes reusable.

Why this should be a community

The best software platforms are not just tools. They are places where knowledge compounds.

GitHub made code reusable. Hugging Face made models easier to discover and build on. Figma made design more collaborative.

Agent work needs a similar place.

When someone publishes a useful way to work with agents, other people should be able to find it, copy it, adapt it, and run it with their own tools.

That public work can also become a portfolio.

In the future, I think many AI power users will be known not only by what they produce, but by the processes they publish. If someone creates the best process for shipping a feature with Claude Code, or researching a market with OpenClaw, or preparing an investor memo, that should be visible and reusable.

This creates a network effect.

More useful public processes attract more users. More users create more useful processes. The best contributors build reputation.

That is the shape I want Epismo to have.

Why I want to build from San Francisco first

I am from Japan, and I know the Japanese AI market well. I have experience, customer understanding, and a network there.

But Epismo is a new category, and new categories should start where the earliest adopters are most concentrated. For agent work, that is San Francisco and the US.

Founders, developers, startups, and power users there are already building new habits around agents. They move fast, share what they learn, and adopt new workflows early.

Japan and other markets can be strong expansion markets later. First, I want to build Epismo with the fastest agent-native users in the world.

How big this can become

Today, most SaaS is priced around human seats.

Agents change that.

One person can run many agents. One team can run thousands of agentic processes. As agents become a normal part of knowledge work, the amount of work done through agents may grow much faster than headcount.

Individuals can start for free. Teams pay for private collaboration and usage. As agent work grows, usage grows with it.

The first market is individuals and startup teams that already use AI heavily. But the long-term market is much larger: every knowledge worker and the agents working with them.

If agents become a normal way to get work done, then the processes behind agent work become valuable infrastructure.

That is the opportunity.

What I believe

“Everyone gets an agent” is too simple.

The future is that people will work through many agents, many tools, and many repeatable processes. Some steps will be done by humans. Some steps will be done by agents. The best teams will be the ones that learn how to combine them well.

But that learning should not stay private.

Today, it is scattered across personal habits, one-off sessions, internal docs, and intuition. One person figures out a better way to work with agents, but the next person still starts from scratch.

I think that knowledge should live in a shared, reusable layer.

That is what Epismo is building: a place where the best ways of working with agents can compound.