16 operators x AI transformation x AI disillusionment

Last week we hosted 16 operators at the CLT Startup House, a dinner I co-hosted with Kristi Straw of Pinnacle Dinners. The dinner was all about AI transformation. We talked about things like, what does it actually take to make AI useful inside a business once the initial excitement wears off? When you get past the novelty and the demos stop feeling like magic.. what happens?

I figured we'd spend a chunk of the night trading stories about where AI transformation got stuck. Here are some of the insights/takeaways..

I've got close friends who have nothing to do with tech. They often look at me like I’m crazy because from where they sit the AI hype doesn’t make sense.

That gap doesn't come from the technology falling short, it comes from people operating with an old mental model of what AI could do a few years ago.

At dinner we talked about how most leader’s entire opinion of AI is based on opening a free version of ChatGPT sometime between 2023-2024, asking it something, watching it fumble, and quietly filing it under “this isn’t that good” or “it’s not ready.”

That was fair at the time, but not now. To put it in perspective, Claude Code didn't even exist until last fall (in AI time, it feels like it’s been years haha).

The thing people wrote off isn't the thing available now.

AI disillusionment isn't a capability problem, it's a people-knowing/understanding what’s possible NOW problem.

Leaders typically fall into one of two buckets right now:

Both groups land in the same spot, and neither one of them have actually tested today's tools against a real piece of their own work.

One operator at the table named the trap for group B pretty perfectly.. “People go chasing one-off tools, this little fix for that little problem, instead of starting from the top and asking what the actual strategy is and which few pillars move the needle.”

The instinct when you feel behind is to grab tools so you look caught up, and your team can smell it. Pick the 1 or 2 outcomes that matter and figure out where AI can move the needle for them.

One workflow that genuinely adds value beats a bunch of unused tools.

The Operator dinner wasn't the only event we had at the house this week that I’ve been chewing on.

The day before, Colin Thornton and Alan Corvallis of Torta Studios came by with sandwiches and jammed on the porch with our members about customer acquisition and retention. Torta helps SaaS companies figure out why people actually adopt a product or quietly slip away from it, and Colin framed adoption really well..

Adoption equals motivation minus friction.

You can reduce the friction all the way down, make a tool dead simple to open, and still get nothing, because if motivation is zero, no amount of friction reduction moves anybody. As Colin put it, Friction is the low-hanging fruit, the part everyone rushes to fix because it's the visible work. The harder question is whether the problem even stack-ranks high enough in someone's day for them to care.

If the pain is real and you make the fix easy to reach, you get a magnetic pull and barely have to push. If the pain is mild, a frictionless tool is just motion. And it's exactly why the panic AI rollout fails. The license was never the hard part. Nobody connected that shiny new tool to a problem the team feels during their daily work, so it just sits there.

An attendee at the dinner was a leader at a mid-sized manufacturer. His approach is to put people close to the work before pointing any technology at it.

When a new technology teammate joins, he has them spend 2 weeks on the manufacturing floor, learning what the company actually makes. Only then do they get to ask where technology helps. They make tangible products, so as he put it, it's a target-rich environment, but you have to understand the work before you can improve it.

That's where motivation actually comes from. Pointing AI at a problem you've never felt and you’ll just get more confusion. Point it at a problem you actually understand and have experienced firsthand, and the whole team can see why it matters.

One operator at the table described watching a COO let an AI model make a commercial decision, feeding in a few data points, taking the output as the answer, and never running it past a single human, when there was no way for the model to check the call on its own.

AI is genuinely strong at the verifiable stuff, the kind where you can run it and see for yourself that it's right, which is exactly why it's so good at code. The discipline is knowing which calls you can let it run and which ones still need a person who can tell when the answer is wrong.

Most of my own building/using AI the last few weeks has gone into one idea, and it lines right up with that verifiable point above. The way almost everyone uses AI is the slowest way there is. You type a request, wait, fix it, ask again, all by hand. You're the engine, and the second you stop pushing, the tool stops.

Using loops completely changes that.

Instead of walking the tool through every step, you give it the goal once and let it run the cycle itself.. plan, do the work, check the result against the goal, fix what's weak, and repeat until it clears the bar. You step out, and the work keeps going.

You don't need to be an engineer to do this.

Open Claude, hand it a goal and a strict set of pass-or-fail criteria, and tell it to score its own work honestly and keep going until every box is green, instead of handing you the first draft that looked close. My one rule.. prove a run by hand before you automate it, and don't force a loop onto work you can't check, or you'll just pay to watch it spin.

Here’s an example prompt you can use in Claude/Gemini/ChatGPT as a non-technical person wanting to test a basic loop to self-verify a plan:

Analyze and create a plan to address for me to review/approve. Then generate 3-5 verification questions that would expose errors in your plan. Answer each verification question independently. Then provide your final revised answer based on the verification.

Your sense of what AI can do might be out of date, and that is what keeps good teams stuck. Take the task you last tried and gave up on (e.g., a board summary, a messy reconciliation, a first-draft proposal, etc.) and run it through Claude (or similar) end to end to see what’s possible now. Bonus points if you connect to several tools and use the mini-loop above to build a plan first. Extra bonus points if you build a reusable skill you can run on a schedule. If you don’t know what that means, ask Claude 😉

Onward & upward 🤘
Drew

P.s. This Operator dinner was a new thing for us, one table, family-style, real conversation, no panels or pitches. We’re planning to make it a regular series at the house. If you're an operator in or around Charlotte and want a seat at the next one, just reply to this and let me know.