What one person can honestly keep running

When you run your business alone, an AI agent that drafts replies or writes code makes extra work tempting. The first delivery looks manageable. Saying yes to ongoing work means keeping it running after the first delivery.

Take a hypothetical request from an existing client: “can your AI just do this as well?” You already prepare their customer updates. The client now wants you to chase suppliers for delivery dates, so the updates contain better information. It sounds like a useful addition. The work continues whenever a supplier is late or changes a date.

Other illustrative situations carry their own follow-up. A new client needs answers after receiving the first piece of work. An extended retainer, an ongoing agreement to provide a service, brings another commitment to revisit when renewal comes round. A promised website feature needs fixing if a later change stops it working. The reply, the renewal and the repair all belong somewhere in the business.

For someone working alone, those commitments sit alongside the next delivery. A good first result can even encourage the client to rely on you more. In the supplier example, the client starts planning around dates you obtained. An unanswered message now affects their plans.

The agent can carry some of the work

An agent can help with recurring tasks when it has the tools and information those tasks need. For example, an agent connected to customer records can use those records in a support conversation. A drafted reply is a useful part of answering a customer.

In the hypothetical supplier job, suppose the agent has access to the correspondence. It drafts a chase for an overdue answer and gathers the replies. Much of the routine writing is covered. The awkward case is a supplier giving different dates in different messages. Someone has to decide which date the client can use.

Chasing an invoice has a similar split. A draft reminder helps; deciding what to say about a disputed bill still needs attention. In software, an agent with access to tests can run checks and use the results. A passing check only tells you about the behaviour examined.

The agent's part needs a record of what happened and someone checking the result. For the supplier work, that means being able to find the message behind the date. If the client calls because a delivery failed, you need enough knowledge to discuss the problem.

The agent makes the yes cheaper; your name stays on the promise.

The number depends on what can go wrong

The amount one person can keep running depends on the failures they can notice and deal with alongside their existing commitments. A count of clients tells you very little about that.

In the supplier example, the client could reasonably ask what happens when nobody replies. They would want to know whether you would spot the missing date before their customer update went out. The client would also want to know who checks a date the agent has extracted from a long email. If the answer is you, checking has to fit alongside the work already promised.

Suppose, in a separate hypothetical situation, a fractional executive, working part time for several businesses, reviews three retainers. One includes regular advice; another involves decisions whenever a problem crops up. The third has grown to include following those decisions through. Three agreements can demand very different levels of attention. Quicker preparation leaves the executive with the calls and decisions each business still expects.

Who notices first matters as much as what breaks. If the client discovers every missed supplier reply, they are doing part of the checking they expected you to provide. A record helps you investigate after a miss. It does not by itself make you notice sooner.

There is judgement in deciding whether another commitment fits. It includes admitting when the extra work only fits if everything else goes to plan.

Some work belongs on the no list

Work belongs on the no list when you cannot provide the continuing attention the client would reasonably expect. Being able to produce the first result does not settle that decision.

And the way I see it, being capable of doing a job is a poor reason to promise an ongoing service you cannot look after. That applies to clients as well as services. A client who needs you available whenever something goes wrong can be a bad fit even when the individual jobs are easy.

For the hypothetical supplier request, assume you can handle the routine chases but cannot reliably follow up conflicting or missing answers. The client needs delivery dates they can plan around. On those terms, the answer is no.

“I won't take on the supplier chasing. I couldn't follow up missing or conflicting dates to the standard you'd need.”

The reason names the part you cannot provide. There is no apology to negotiate past, and no suggestion that a quicker draft would change the answer. The client can decide who else should take the work on.

A feature can belong on the same list. If adding it means supporting a connection to another system you cannot keep checking, declining the feature is a reasonable decision before anybody starts relying on it.

The next yes includes the follow-up

The next yes needs room for the work that continues once the first result is in use. For the hypothetical client, declining supplier chasing leaves an existing service to look after: preparing customer updates from information the client supplies.

Another request may fit. It needs considering alongside the unanswered messages and fixes already due, including the ones an agent has helped prepare. A clear inbox after sending replies does not mean every issue in those replies is resolved.

Before the next yes, picture the client returning because the result they rely on has stopped working. The extra commitment has to include dealing with that return.