The Wild West of AI, Part 2: How to Navigate It

The people pulling ahead in the Wild West of AI aren't the ones who read the most. They're not refreshing their newsletter feed on Sunday night or rewatching the DevDay recap to catch what they missed. They picked one problem, ignored most of the noise, and built something with it.

That's what we covered in Part 1 of this series: why "keeping up" quietly became impossible. This one covers what to do instead: focusing on what you're actually trying to solve and putting structure around it. In practice, that looks like this:

1. Start with a problem, not a tool.
‍Stop asking, “how can we use AI?” and start asking “what is the most expensive, repetitive, mentally draining problem in this department?” Then, and only then, ask whether AI can solve it. The teams creating real value right now aren’t the ones using the newest model. They're the ones who picked one problem worth solving and refused to get distracted. A tool without a problem is just a subscription.

2. Judge every tool by the questions that don’t change.
‍The “best model” changes monthly. Your evaluation criteria shouldn’t change at all. Before anything touches real work, it answers the same boring questions. Where does our data go? Who can access it? Does it meet the compliance requirements for your industry, whether that's HIPAA, SOC 2, GDPR, or whatever alphabet soup applies to you? Can we swap it out in six months without burning everything down? Boring questions are how you stay standing when the exciting tool gets acquired, deprecated, or breached.

3. Write it down so it transfers.
‍A great prompt living in someone’s head helps exactly one person. A documented workflow, with the use case, the tool, the guardrails, and the result written down, is something another team can pick up and run. That’s the entire difference between a company experimenting with AI and a company creating value with it. It’s also the first real fix for the doing-this-alone problem: the moment your work is written down, other people can finally tell you whether it’s any good.

4. Measure like a skeptic.
‍“The team loves it” is not a metric. Hours saved, error rate, cycle time, cost per output. Pick your numbers before the pilot starts, or the pilot never ends. And remember, nobody is going to hand you a grade for this. Your metrics are how you gauge actual success. Measurement is the feedback loop your company never gave you.

5. Stop learning in secret.
‍You are not the only person at your company teaching yourself AI on the weekend and wondering if you’re doing it right. Say what you're trying out loud. Start the channel, show a coworker your workflow, ask your manager what the actual policy is (asking is usually what forces one to get written). You can’t find out whether you’re doing it correctly until someone else can see what you’re doing.

6. Put yourself on an information diet.
‍Pick two or three sources you actually trust. Give them thirty minutes, once or twice a week, on your calendar. Let everything else go. If a release truly matters, it will still matter at your next check-in. And if a teammate mentions a model you’ve never heard of? That’s ok, they just gave you up-to-date information on AI you didn’t have to spend mental cycles finding on your own.

The Wild West Doesn’t Settle Down, But You Can.

The noise is not going to stop. There will be anew best model next month, and another one the month after that. You can’t control that. What you can control is what your team builds in the meantime. You can make a short list of real problems, solved with tools that pass your security and compliance questions, written down so the value transfers, measured so you know it’s real, and shared so nobody is figuring this out alone anymore.

The developer you eat lunch with is always going to know about some model you don’t, and that's ok. Feeling “caught up” with AI doesn’t matter if you can’t point to actual, repeatable value from AI workflows and show exactly how it was built as well as how to build it again.

That’s how you survive the Wild West of AI. Not by outrunning it. By outlasting it.