7 min read

How founders should use AI agents without handing over the business

The question is not whether to use AI. It is what you let it touch.

Founders are being told two contradictory things about AI: that it will run the business for them, and that handing it the keys is reckless. Both contain something true. The useful question is not whether to use AI agents. It is what you let them touch.

Propose, do not act

The reliable pattern today is simple to state: agents propose, the founder decides. An AI agent can read what is happening in your business and draft the next move — the weekly plan, the delegation, the follow-up, the flag on a bottleneck — and then route it to a person to approve, edit, or reject. What it does not do is act on its own: no email sent, no money spent, no commitment made, without a human decision in between.

This is not timidity. It is the difference between a tool that multiplies your judgement and one that quietly replaces it with its own.

The boundary that keeps you in control

The safe shape is a clear asymmetry of access. Give an agent the ability to read a defined model of your business and to suggest into a review queue. Do not give it write access to your tools, your accounts, or your inbox. You issue that access and you revoke it. The agent can see and recommend; it cannot reach out and change the world. That single boundary is what lets you gain the leverage without inheriting a new, less accountable bottleneck.

What this looks like in practice

Concretely: an agent that can read your operating model proposes that a recurring decision now has enough history to be delegated, and drafts the handover. It notices a meeting that has lost its purpose and suggests removing it. It prepares the weekly brief from the week's signals. Each arrives as a recommendation you act on in seconds — or do not. The work of thinking is shared; the authority to commit stays with you.

Why this matters more as agents get capable

As agents grow more capable, the temptation to let them act unattended will grow with them, and so will the cost of a confident mistake. The founders who do best will not be the ones who hand over the most. They will be the ones who built a clear model of their business for an agent to reason over, and kept the decisions that matter on their own desk. The goal has not changed: a business that depends less on you, run on judgement that is still yours.

Questions

Can AI run my business operations?

It can assist with most of them and should run none of them unattended. The reliable pattern today is propose-and-approve: an AI agent drafts the brief, suggests the delegation, flags the bottleneck — and a person decides. Letting an agent act, send, or spend on its own trades the founder's bottleneck for a less accountable one.

Is it safe to connect AI agents to my business data?

It can be, with the right boundary. Give an agent read access to a defined operating model and a proposal channel into a review queue, not write access to your tools, money, or inbox. Issue and revoke that access yourself. The safety comes from the agent being able to read and suggest but never to act without a human decision.

Start with a model worth connecting

A three-minute diagnostic that begins the operating model an AI agent can later read — and propose against, never act on.

Start with a model worth connecting