# What's the Ideal Company-Brain Harness?

> Everyone's defining the ideal coding agent. The company-brain harness people actually want knows the company, clears urgent work, stays multiplayer -- and stays open to the agents they already use.
- **Author**: Ferran Sulaiman
- **Published**: 2026-09-26
- **URL**: https://ferran.sh/writing/whats-the-ideal-company-brain-harness

---

Everyone's defining the ideal coding agent. What's the ideal company-brain agent? What's the ideal company harness?

People often conflate these terms together, so let's clear the air early.

**Company brain** is the knowledge layer -- what's going on now, what happened before, the stuff only old employees remember, the processes people only carry in their heads, etc. Historical plus current. That's the brain.

**Company brain agent** is the AI assistant sitting on top of that brain. The one with arms and legs that can actually do things for you.

**Company harness** is the environment that gives your agent its arms and legs. Without the harness, your agent just spits out words.

When a company wants an agent for the entire team, it's usually to tend to the same wound: you want to spend time on what's important but you keep getting pulled into what's urgent.

When that trade goes wrong for long enough, growth slows. You've seen it. I've seen it. Companies either stall or disappear over time.

So the job of the agent is not to get you to "inbox zero". Its job is to take up the urgent work that doesn't need you so that you can always focus on what's important.

On top of that, it's important to note that not everybody wants to use the company's own agent. Most people seem to have their own elaborate setups they prefer using over any other set of new tooling that you can give them.

But what they do find valuable is the **company brain**.

If your company brain agent can reach historical archives and current data and surface what matters and act on it -- great. Your non-tech-savvy teammates have an equally capable assistant that they can use to be as productive as you are.

But there are people who want to be able to plug in the company brain into their harness of choice (think Claude Cowork, ChatGPT Work, heck even Copilot, for that matter.)

Could those agents consume the company brain that you've built?

That portable-brain requirement sits on top of a short list of what a true company brain harness should consist of.

**1. The agent in the harness has to know the company**

Pull from anywhere the company stores truth. Enterprise search is now easy. **Enterprise reasoning**, however, is an unsolved problem.

**2. The agent has to act on urgent-not-important work**

"Inbox zero" is an example. Can it do follow ups? Can it take care of a routine API key rotation? Can the agent free up your mindshare to focus on what actually matters?

**3. The harness has to make the agent proactive**

Sure, the agent responds when asked. Does something when told. But you want it to do more. You want your agent to tell you what's not done, flag when you're focusing on the wrong things and push you in the right direction (on an individual level).

**4. The harness has to be multiplayer**

If you're serious about agentic workflows, you can't wall each person to their own agent. You should be able to allow another user to interact with your agent. You should be able to tag another user's agent to review your work with their context. Maybe you, your teammate and their agent is the dream team you've always wanted.

That's what I'd buy.

I don't care about the perfect company agent.

A company brain that other agents can use.
A harness that makes my agent proactive and stays multiplayer.

Everything else is a demo.

---
- [All articles](https://ferran.sh/writing)