You own this now

I was waiting for a retailer to resolve a small customer-service issue. I had followed up a couple of times, heard nothing for a while, and then continued with my life.

The problem was not the effort required to write another email. That took a few minutes at most. The problem was that the issue kept returning to me at random. I would remember it days later, send another note, then forget about it until the next time it resurfaced.

It had become another mental tab: not important enough to deserve sustained attention, but not resolved enough to close.

So I handed the follow-up to an AI agent. The instruction was deliberately simple: keep sending regular email follow-ups until a response arrives, then stop.

That felt different from asking AI to draft a message. Drafting still leaves the responsibility with me. I have to remember the issue, ask for the email, review it, send it, and repeat the whole sequence if nobody replies.

This time I was delegating the responsibility itself. The task was no longer ‘write a follow-up.’ It was ‘keep this moving until the situation changes.’

You own this now. Come back to me when something changes.

In literal time, the automation probably saved me very little. Even several follow-up emails would not have amounted to much actual work.

But time saved is a poor measure of this particular value. The agent removed the need to remember. I could stop carrying the unresolved issue in the background because something else was now responsible for checking, acting and stopping at the right moment.

That is a form of cognitive relief. The benefit is not only fewer clicks; it is being able to trust that a small obligation will not disappear merely because my attention moved elsewhere.

The workflow worked because the responsibility was narrow. There was one existing issue, one conversation to follow, a defined cadence and a clear stopping condition: a response had arrived.

That last part matters. ‘Keep following up’ without limits is not delegation; it is a recipe for becoming the most irritating person in somebody else’s inbox. An agent needs to know not only what action it may take, but when the job is complete and when an exception should return to a human.

I am increasingly convinced that this is where practical agent design begins: not with maximum autonomy, but with a well-shaped responsibility. The clearer the boundary, the easier it is to trust the system inside it.

• A specific task and recipient
• A sensible follow-up cadence
• A visible record of what was sent
• A clear stop condition
• Human escalation when the situation changes

The most visible examples of agentic AI tend to be ambitious: agents that research a market, operate across several tools or attempt an entire role. Those examples are useful, but they can make the category feel more distant than it is.

This use case was almost aggressively ordinary. A small issue needed polite persistence. The agent watched for a change, repeated a bounded action and stopped when its job was done.

That is where I am seeing value today. Day-to-day life is full of responsibilities that are easy to perform once and annoying to keep carrying: chase a response, check whether something changed, or keep a process moving until the next decision is required.

The opportunity is not to hand every judgement to an AI system. It is to decide which responsibilities genuinely require me, and which ones I should be able to hand off with confidence.