Welcome back. Issue two.

This one is about the question I get more than any other, from owners at meetups, from prospects on first calls, from friends of friends at barbecues: "Where do we even start with AI?" The answer is simpler than the industry wants it to be. Let's get into it.

Quick note before we dig in: I'm on the Founder Series panel at Grow House in Denver tomorrow afternoon, Artificial Intelligence and the New Frontier. If you're local, come say hi.

The pattern we keep seeing

The companies that get burned on AI all start the same way. They pick a technology, then go hunting for a problem to aim it at. Someone sees a demo, a budget gets approved, and six months later there's a chatbot nobody uses and a board update that says "we're doing AI."

The companies that win invert it. They start with the pain. The inbox someone dreads opening. The report that eats every Friday afternoon. The phone that rings into voicemail while the whole crew is in the field.

Here's why the order matters so much. When you start with a technology, success is undefined, so the project can't end, it can only fade. When you start with a pain, success is obvious: the pain is gone or it isn't. The team can feel it. The books can show it. And a proven first win does something no strategy deck can do: it earns the trust for the second project, and the second one funds the third.

Fix one. Prove it. Then earn the next one. That's how you buy AI without getting sold AI.

From the field: the mapping problem

A private equity team had a chore attached to every deal they evaluated. To compare a target company's books to their standard model, someone had to map the target's chart of accounts, hundreds of general ledger accounts, matched line by line, by hand. 40 to 80 hours per deal. Skilled analysts, doing matching work, deal after deal.

Nobody called it an AI opportunity. They called it Tuesday.

We built an agent that does the mapping and, crucially, shows its confidence on every line. High-confidence matches sail through. Low-confidence ones route to a human. The analysts stopped grinding and started reviewing. The same work now takes under 4 hours per deal.

Notice what made this the right first project: repetitive, rule-based, every single deal, and everyone involved hated it. That's what your first project should feel like.

From the field, part two: the phone that stopped ringing into the void

A different flavor of the same lesson. A service company with crews in the field all day was missing calls constantly, and every missed call was a customer dialing the next name on the list. Their first instinct was a bigger office team. Their actual problem was a coverage gap, not a headcount gap.

So instead of a hire, they got a voice agent. She answers every call, asks the same qualifying questions their best office person would ask, books the estimate on the real calendar, and texts the customer a confirmation. Nights, weekends, holidays.

Two projects, two industries, one pattern: neither started with "we need AI." Both started with a pain everyone already complained about. The technology was the last decision, not the first.

Try this tomorrow

The Tuesday walk. It costs nothing and takes an hour. Walk your own office on a normal Tuesday with a notepad and write down every task someone does that follows a checklist.

Two rules. Watch the work, don't ask about it in the abstract, because the process people describe in meetings is never quite the process they run. And note the groans: the task someone sighs before starting is telling you something no analysis can.

Then price one item with three numbers: times per month, minutes each, loaded hourly cost. Multiply. A 10-minute task done 300 times a month at $50 an hour is $2,500 a month, $30,000 a year, forever, and nobody ever approved that spend. It just accumulated.

That list, priced, is your AI roadmap in priority order.

Questions from readers

Two that came in after issue one, answered the way I'd answer over coffee.

"This all sounds great for bigger companies. We're 12 people. Does any of it apply?" Backwards, honestly. The most aggressive AI wins we see are in 10-to-50-person companies, because the owner can decide on a Tuesday and ship by month-end. No committee, no procurement, no pilot review board. Small is an advantage here, maybe for the first time in software history.

"How do I know if a task is actually automatable or if I'm dreaming?" Three tests. Is it repetitive? Is it rule-based (could you write the checklist)? Is it high volume? Hit all three and it's a candidate. Miss one and it's probably a judgment task, which means it's the part of the job a human should keep.

Got a question for issue three? Just reply to this email. I read every one.

What we're seeing

Budget questions are starting early: companies are already asking how AI fits 2027 planning, months ahead of season. Voice is crossing the believability line: conversations that started with skepticism a year ago now start with "my buddy's company has one." And the ban era is quietly ending: fewer companies asking how to block tools, more asking how to give their team a sanctioned one.

One thing worth knowing

Here's a pattern worth naming before budget season: the companies that started with one small workflow this year will walk into 2027 with proof, believers, and a funded second project. The companies "still evaluating" will walk in with a kickoff meeting. Same calendar, completely different year. The gap between them isn't spend or talent. It's a start date that already happened.

If you want to skip the guesswork and find your step one in 30 minutes, grab a slot: https://calendly.com/d/dv4k-67p-3bd

One small ask

If this issue was useful, forward it to one business owner who'd like it. That's how the Brief grows.

Until next time,

Shawn Kercher
Co-Founder and CTO, HummingAgent AI