The Jarvis Idea, Minus the Suit

Everybody who saw those films wanted the assistant more than the suit. Not because it answered questions, but because it noticed things, made decisions and got on with them.

For a long time that was purely fiction. What we had were systems that followed rules, and rules only cover the situations somebody thought of in advance.

Agents are the shift, and both Claude and ChatGPT can now plan a task and use tools to carry it out. Give one a goal rather than a script, and it works out the steps, uses the tools it has, and adapts when a step fails. That is a different kind of thing from an automation, and it is close enough to reach now.

We are careful not to oversell it. Today’s agents are more capable at planning than at judgement, and they need supervision on anything that matters. The right posture is a promising junior colleague, not an infallible one.

Still, the direction is clear. Less asking, more anticipating. Less operating the tool, more agreeing on the outcome.

Building an agent that actually runs is where the AI Innovator Program finishes, and people leave with one working rather than a diagram of one.