You can't start on third base

Written by Jordan Saunders | Aug 18, 2026, 6:15:35 PM

Agent technology is moving faster than anything I have watched in my career. And most companies I talk to are standing still anyway.

I do not think those companies are wrong to hesitate. I think they are reading their situation correctly and drawing the wrong conclusion from it. That is what this is about.

The Pitch You Are Getting from the Internet

Here is the pitch you are getting from the internet right now. Stand up a company-wide knowledge base that answers every question, automates half the back office, and lets you replace departments. That is third base. The demos are real, some of them anyway, but they all skip the part where somebody had to get to first.

Meanwhile your hesitation is rational, and I want to say that plainly because almost nobody selling AI will. Your business works. You have customers to serve this week and payroll on Friday, and keeping that machine running is worth more than any pilot. Change too fast and your workforce cannot function — and here is the part that gets missed: they will not adopt the change either. An automation nobody adopts does not compound. It sits in the middle of a working process like a stalled car.

And the big swing has a second problem beyond adoption, which is scope. A company-wide AI initiative is enormous, the level of effort is nearly impossible to estimate, and once you have bitten it off you cannot steer it. This is classic change management.

The Answer Is a First Win

So the answer is not zero and it is not third base. It is a first win.

A first win is anything that automates a real piece of work and saves a real person time. That is the whole bar. If one person gets hours back and actually uses the thing, it counts.

Here are two of ours, so you can see the scale I mean.

Meeting transcripts that maintain themselves. We use meeting transcripts to keep the notes on our internal initiatives up to date, which means the record maintains itself and nobody spends Friday reconstructing what was decided.

AI agents screening fraudulent applicants. We run AI agents in Ashby, our hiring system, screening out fraudulent applicants before a human ever reads them. That one cut our time to hire in half.

Neither of these would make a keynote. Both of them are running today, saving actual hours, and nobody on the team had to relearn their job to get the benefit.

Why Small Works

I run a custom software firm, so weigh my incentive. But notice the advice is to start smaller than anyone would hire us for.

Small works for a reason.

A small win teaches you how to evaluate this stuff — what the effort really looks like, where adoption gets sticky — before the stakes are high.

It asks almost nothing of the people around it, so it actually gets used.

It earns you the next one. Automations compound, but only in companies where the last one got adopted. That is how you get to third base, one base at a time.

The companies already there did it the same way.

What to Do With This in the Next Week

Draw a 2x2. Horizontal axis: how many people have to change how they work — few on the left, many on the right. Vertical axis: hours saved per week on work someone already does — low at the bottom, high at the top.

Now plot every AI idea floating around your company.

The department-replacement stuff lands far right — lots of people changing how they work. Ignore it for now.

The clever ideas that save almost no time land at the bottom. Ignore those too, even the fun ones.

What you want is the top left corner: real hours saved, on work that already happens, where almost nobody has to change anything.

Something is sitting in that corner. Maybe it is a report someone assembles by hand every week, or a screening step, or notes nobody keeps current. It will look too small to matter. That is what a first win looks like.

Do that one first, and just the one.

View full post