Your AI Should Be a Workhorse

Until recently, getting real value out of AI meant having a conversation. You open a chat, you type, it answers, you steer it to the next step. That back-and-forth felt like the whole point. It is already the old way.

A conversation partner needs you present for every exchange. You are the one moving it forward, one prompt at a time. The work only goes as fast as you can type and think and respond. You are in the loop, always, whether your judgment is needed at that moment or not.

There is a different way to work, and it changes the economics completely.

The newsletter that drafts itself on a Monday

I send a newsletter every week. For a long time it took a two-hour block. Open a chat, pull up the week's notes, prompt the model, draft a section, refine it, draft the next one. Good work, but slow, and it needed me in the chair for all of it.

Now a skill and a scheduled agent do the first pass. Once a week it reads through the meeting transcripts that have landed in my notes, finds the themes I keep coming back to across different calls, builds the thread, and drafts the whole email in the structure I have trained it on. It produces three options and drops them into my to-do list every Monday morning.

Two hours became about twenty minutes. Not because the writing got faster. Because I stopped doing the parts that never needed me. I show up at the end, read what it found, and decide. The judgment is mine. The fetching and the sorting and the first draft are not.

That is the shift. From AI as a conversation partner to AI as a workhorse.

Conversation partner, or workhorse

A conversation partner waits for you. It is sharp in the moment and idle the rest of the time. The upside is capped by your own pace, because nothing happens unless you are there to prompt the next move.

A workhorse runs on a schedule. It does the regular, repeatable work on its own and hands you back something to judge. You spend your attention where it counts and nowhere else.

The same machine can play either role. The difference is not the model. It is whether you have given the work a heartbeat, a time it runs without being asked.

The real question is where you are the bottleneck

AI is faster than you at the repeatable parts. You are faster than no one and slower than the machine at the things that repeat. So for any recurring task, the question is not "how do I prompt this better." It is "where in this loop does my judgment actually add something, and where am I just the bottleneck."

Find the judgment point. That is the part only you can do: the taste, the decision, the relationship, the call on whether this is true and ready to go out under your name. Keep yourself firmly there.

Then look at everything around it. The gathering, the sorting, the formatting, the first draft, the chasing. That is the repeatable work, and that is what the workhorse is for.

This holds well beyond AI. Most of the time we feel stretched, the problem is not that there is too much work. It is that we have wedged ourselves into the middle of loops where our judgment was never required.

The failure mode to watch

There is a way to get this wrong. You remove yourself from a point where your judgment did matter. You let the agent run, and you ship its eighty percent as if it were the finished hundred.

That is the trap. The workhorse does the draft. It does not do the deciding. My newsletter still gets twenty minutes of me reading every line, because the call on what is true and what is mine to say cannot be handed off. Automate the fetching. Keep the discernment.

Review is the loop you keep.

What to do with this

Pick one task you repeat every week. The newsletter. The content round-up. The client follow-up. The report you rebuild from scratch every Monday.

Now split it in two. Write down the part that genuinely needs your judgment, and write down everything else. Be honest about how small the judgment part actually is. For most weekly tasks it is the last ten minutes, not the first two hours.

Then build the workhorse for the rest. Give it the structure you would use, give it a time to run, and let it hand you a draft to react to. Keep yourself at the judgment point. Step out of everywhere you were only ever the bottleneck.

The conversation was a good way to start. It is not the way to scale.