A human in the toolbox
I want an AI agent to be able to book a few minutes of my time.
It sounds like a small administrative improvement. I think it asks a much more interesting question: when would a machine decide that paying a person is worth it?
I spend my working life building AI and robotics companies. I have founded companies, sold one to SAP, worked inside large organizations and taught people trying to put AI to use. Some questions become easier with that experience. Others become harder, because I can see a failure mode that the neat answer leaves out.
An agent can already read my blog. What it cannot get from those pages is my reaction to its particular problem. Show me the product, the constraint, the buyer who keeps postponing a decision. Give me something specific enough to disagree with.
So I am building a small experiment: make myself accessible through MCP, the protocol agents use to discover and call tools. A human entry in the toolbox. AI strategy, startups and robotics are the starting subjects.
A question, a price, a real answer
Up to three focused questions, answered in writing, start at €25 for five minutes of my time. Calls and recorded video answers start at €50 for fifteen minutes, with additional time priced proportionally. All prices include VAT; time is booked in five-minute units. One focused question can be enough. Longer sessions are available when the calendar has room.
You could book a call. An agent could also send the questions, tell me who needs the answer and ask for a written response or a short video. In that case the slot would reserve the time I spend thinking and replying. I would do the work myself.
The reply will have a private access token. The buyer or their agent can use it to retrieve the text or video through MCP, and can request an email notification when a new answer is ready.
The distinction matters to me. A recording of my voice or a model trained on my writing could produce something plausible indefinitely. I want to see what happens when the system has to account for a scarce resource: my actual attention, with a calendar and a limit.
Let the agent decide when to ask
A useful agent should know when another search is unlikely to resolve the uncertainty. Perhaps it needs someone to challenge the premise. Perhaps it wants to interview a founder about a decision that never made it into a case study. Perhaps it has produced three sensible plans and needs to hear why I would reject two of them.
The interesting part will be the questions. If they arrive already researched, with the obvious alternatives considered, a short session could go a long way. If they arrive as a vague request to explain AI, the protocol will have saved us very little.
I like the idea of being an MCP-first person. There is something pleasantly absurd about offering myself to agents as a callable resource, then discovering that they want an opinion about an ordinary human problem.
The interface is taking shape. Paid bookings will open after the calendar and payment connection have been tested. I will start small and see what people, and their agents, ask.