The Free 2-Week Proof Run

Pick one job your team does by hand. In two weeks, you'll know if AI can do it.

I build it. It runs on your real work for two weeks, next to your team. Then we put its numbers beside theirs. You keep the build and the numbers either way.

  • Two weeks. Your real work, not a demo.

  • Nothing upfront. You pay me only if it works and you want to keep it.

  • You keep it. The build and the numbers are yours, even if the answer is no.

Tell me what your team does by hand

Tell me what your team does by hand

Two or three sentences. Not sure which job? Tell me what the business does and I'll pick one. I'll tell you straight whether it's a fit — including when it isn't.

I read every one of these myself. No list, no drip, no pitch deck.

Is this you?

This is for you if

  • You own the business, or you're about to. It's your money and your call.
  • Your team does the same steps all day — the same questions, the same forms, the same calls.
  • You already use AI and pay for it. You just don't know where it goes in your business.
  • You can point at the number this process moves.

This isn't for you if

  • You'd need convincing that anything is wrong. I don't do convincing.
  • The work changes every time and there's no pattern in it.
  • You want someone to join your dev team. That's a different job and I'm not it.

Two Weeks, One Process

What a Proof Run is

One process. Real volume. A number at the end that you own.

What you get

  • A working system that does the job, running on your real work.
  • One page at the end: what it handled, what your team handled, side by side on the same work — walked through with you.
  • The build itself. It runs on your own AI account, so you see and cap every penny it costs.
  • A straight answer on whether it's worth keeping — including "no."

What I need from you

  • A few hours with the person who actually does the work. Not the manual — the real version, with the judgement calls in it.
  • Access to wherever the work happens.
  • One person I can ask questions during the two weeks.
  • A budget for your own AI bill. It runs on your account, so the usage is yours — small, visible, and capped wherever you want it.

What it costs

  • Nothing upfront. No service fee.
  • The build is free right now. I'm using these to pick my next product, and I take one at a time.
  • The running system uses your own AI account — the model provider bills you directly, so you do need a budget for that.
  • If it works and you keep it, we talk about money then. Not now. Here's what that looks like.

What you're probably thinking

“Our process is messier than that.”

You don't have to believe me. In two weeks you'll have a number instead of an opinion, and it cost you nothing to get it.

“AI projects take six months.”

Two weeks to real work. Faster if you already have a prompt that half-works.

“We don't have the bandwidth for a project.”

A few hours from the person who does the job. I do the rest.

“Even if it works, so what?”

If we can't name the cost line or the revenue line before we start, I'll tell you it's not a fit and we'll both save the two weeks.

Proof

The numbers so far

+7 pts

A booking script, run by text, beat the phone team

A national direct-sales recruiting operation ran the same booking script by phone. The agent I built ran that script by text across about 2,500 recruits and booked around 74% — roughly 7 points better than the human team.

Read the write-up →

17

Seventeen buyers out of a list somebody gave up on

523 dead leads in, 17 qualified buyers back out, about $200 of model cost to find them. The team closed about $2,000 of business from those; the agent found them, the people closed them.

Read the write-up →

45 min

An analyst's quarter of research, done before lunch

Building a prospect list of institutions that could buy an investment strategy used to mean a $16,500-a-year database or an analyst and a few months. From public SEC and IRS filings, in 45 minutes of compute and zero dollars of data: 809 institutions and 4,697 named contacts, each tied to the filing it came from. An independent check scored the list 94% precise and caught all 10 planted fakes. Leads, not engaged leads — the write-up is honest about that.

Read the write-up →

6

Two years of support tickets, boiled down to six problems

A team fixing problems reactively, with no view of what was actually driving ticket volume. A pipeline read thousands of support conversations, two years of them, summarized each, and clustered them into the six product problems behind most of the tickets.

Read the write-up →

20¢

One conversation, priced to the penny

Nobody knew what one agent conversation actually cost. Two weeks of billing logs, 42,405 requests, answered it: about 20 cents per recruit conversation, roughly $40 on a weekday. The same logs showed the real ceiling wasn't speed, it was the spending cap, and that one line in the prompt was defeating the provider's cache and costing about half the bill.

Read the write-up →

Neither number is a promise. What I promise is that you get one of your own — on your work, in two weeks. Other things I've built →

What does your team do by hand?

Tell me the process and roughly what it costs you. Not sure which one? Tell me what the business does. I'll tell you straight whether it's a fit.

Tell me what your team does by hand

Remote · US Central Time · One proof run at a time