AI consulting for companies that want to run on agents

Most companies that hire an AI consultant get the same thing. A workshop, a maturity score, a list of tools, and a roadmap that nobody owns three weeks later. The company runs exactly the way it did before, with a few more subscriptions.

That is not what the technology is for anymore. The current generation of AI does not answer questions about the work. It does the work. The question is not which tools to buy. It is what has to be true about your company for an agent to be trusted with a real job, and in what order to make it true.

That is what I do.

Who I am

I'm Ethan Rogers. I build companies that run on agents, and the infrastructure they run on.

I own a piece of Plastix, a marketing agency that is being rebuilt to run on agents. I built INKRALLY, a canvas art brand where GPT-6 now runs the whole business. I build the tools I wish existed, like Billr, which turns agent work into invoices a client will pay, and I release most of what I build as open source.

The open source part is how you can check my work before you ever talk to me. Harvey, an autonomous AI sales agent, has over 2,000 stars on GitHub. JARVIS, a voice assistant for your Mac, and Codex on Crack, which coordinates several models from one conversation, have hundreds each. Everything is listed on resources.

What an engagement looks like

I take a few engagements a year. Each one has a fixed scope, written success criteria, and an executive sponsor on your side who can make decisions. There are two shapes.

A consulting sprint

A short, focused engagement that ends with decisions, not a deck.

• Where the work actually goes. We map how work moves through the business today, by talking to the people who do it, not the org chart.
• What to give agents first. Ranked by value and by how ready each process is. Some of your most valuable work is not ready, and I will tell you why.
• What has to be written down. The processes, boundaries and approvals an agent needs, in the words your team uses.
• A plan your team can execute. Specific enough that you could build it without me if you wanted to.

A full build

The sprint, then we build it. Agents doing real jobs in your systems, with the parts most projects skip: who each agent is, what it may do alone, how it reaches a person when it needs one, what it costs, and a record of what it did. Your team learns to run and change it as we go, because the goal is that you do not need me afterward.

How I think about it

Execution is becoming free. Direction is not.

The cost of producing work has collapsed. Code, copy, analysis, support, research. But a company was never just the work it produces. It is the apparatus around the work: who does what, who is allowed to do what, how work reaches the right place, what it costs, and what happened. For people, that apparatus took a century to build. For agents, none of it exists yet.

So I work in a specific order. Every person works through AI directly first, because that is how the real processes surface. Then the company gets written down and made reachable. Only then do agents run loops on their own, with the identity, authority, direction, accounting and record that let you trust them. Skip a step and the project dies for "unclear business value," which is how most agent initiatives end. I wrote the long version in How Companies Will Be Run by AI Agents.

Who this is for

A good fit: a service business with a team, real revenue, and an owner who wants the work to run without them. Agencies, professional services, operations-heavy businesses, and teams that already use AI every day but have not changed how the company runs.

Not a fit: hosting, ongoing maintenance, staff augmentation, or a company looking for someone to approve a tool purchase. If you want a chatbot on your website, you do not need me.

What you leave with

• Agents doing real work, in your systems, that you own outright
• The company written down, in documents both people and agents can work from
• Clear authority: what agents do alone, what needs approval, what is forbidden
• Costs you can see, per task and per agent
• A team that can run it, change it, and build the next one

If that is what you want, tell me what you run and what you want to change. I read every message and reply within two working days.