Ensemble's first alpha customer, and what an AI agent is actually for.

Updated 30 September 2026: an earlier version described this as a paying customer and described deployment options as built. Ensemble is in development; this version corrects that.

This week, Ensemble took on its first alpha customer. An alpha customer works with us while the product is still being built. We are building an agent with them that helps with their day-to-day work.

That is the post. The rest is context.

I want to start there because every AI launch post I have read in the last year has buried the part of the story that actually matters. The interesting question is not which model an agent uses, or how clever the orchestration is, or how fast the demo runs. The interesting question is whether the agent does real work for someone other than the person who built it. And for an entire class of buyer, a second question comes right behind it: where does the data go?

Why we are building Ensemble

Two patterns built up in our network over the last year.

The first is the operator who wants AI agents inside their company and does not know where to start. They have heard the names. Agent frameworks, MCP, the latest open-source projects. None of it adds up to a thing they can actually use. The distance between “AI agents are powerful” and “we run an AI agent that does our work” turns out to be wider than the marketing suggests.

The second is the operator who has crossed that distance. Founders and engineers who have assembled agent stacks that genuinely work. They run on a laptop, a terminal, a Python script, sometimes a Docker container on a server they own. The agents do useful things. They draft, they summarise, they pull from a CRM, they reason over a knowledge base.

Then a real conversation starts, or a real client needs an answer, or someone on the team needs to actually use the thing. And the agent stays in the terminal. The script never makes it into the room where the work happens.

I started calling that the self-hosting graveyard. Capable agents, real engineering behind them, and no surface for anyone other than the builder to use them on.

We are building Ensemble to solve both. Two things. Build the agent, the part that knows your work, your data, the way decisions actually get made in your firm. And get the agent out of the terminal and in front of the people who need it, not just the person who built it.

What “alpha customer” means

I want to be careful here, because launch posts stretch words like “customer” a long way. So plainly.

An alpha customer is working with us on a real use case while Ensemble is in development. They get an agent built around their work. We get to learn, early and honestly, what an agent has to do to be useful to someone who did not build it. It is not general availability, and it is not a logo on a slide.

I am not going to name them. That part stays with them.

Why where the data goes still matters

For a lot of buyers, regulated or simply careful, where the data goes is not a detail to sort out later. It decides whether an agent is usable at all.

That is a principle we are building toward, not a finished product we are announcing. How Ensemble will be offered is still being worked out, and we will say more when it is real.

What this signals

The playbook is the same one we have run from the start. Build the version the most demanding buyers can actually use, then scale outward. Demanding sometimes means regulated. It also means private, careful, principled about where data goes. That is how I expect the next generation of AI agents to get adopted. Not by lowering the bar on data control. By raising the bar on what a product is willing to ship.

Ensemble is not open to the public yet. If an agent that does real work, where your data stays under your control, is the thing you are waiting for, that is exactly what we are building.