The demo agent and the production agent are different species

🤖 The demo agent and the production agent are different species.

Manus has done something useful this month: it has put autonomous agents on every leadership agenda. Invite codes are reportedly reselling for thousands of dollars, the demo videos are everywhere, some are calling it a second DeepSeek moment — and somewhere in your organisation a senior stakeholder has already asked, "why don't we have this?"

Having shipped agentic systems into an enterprise, here is the distinction I keep drawing for leadership teams:

A demo agent is optimised for the happy path. It books the trip, builds the website, finishes the task — once, on camera, with nobody's data at stake.

A production agent lives in a different world. It touches CRM records that sales teams depend on. It reads documents with client names in them. It produces output that someone will act on in a real pursuit. In that world, the hard questions are not "can it do the task" but:

What is it allowed to touch, and what is it explicitly not?
When it is uncertain, does it guess or does it escalate to a human?
Can we trace why it produced what it produced?
When it fails — and it will — what is the blast radius?

My working rule: autonomy is a dial, not a switch. You earn each notch with evidence — grounded outputs, regression discipline, human gates at the decisions that matter. The viral demos start at full autonomy and work backwards to trust. Enterprises have to start at trust and work forwards to autonomy.

None of this is an argument against agents. It is an argument for sequencing. The organisations that will win this wave are not the ones that deploy the most autonomous agent first — they are the ones whose users still trust the agent in month six.

Here is the question I would put to any leadership team right now: what evidence would you need to see before letting an agent act — not draft, act — on something client-facing? If you cannot answer that, you are not ready for the demo you just watched.

Reading: https://techcrunch.com/2025/03/09/manus-probably-isnt-chinas-second-deepseek-moment/