I've been in technology for 20 years. I've watched the hype cycles. Client-server was going to change everything. Then SOA. Then the cloud. Then Big Data. Then blockchain (remember blockchain?). Then machine learning. Now AI.

Some of those things genuinely changed everything. Some didn't. Most were real technologies that got buried under so much hype that it became impossible to have a sensible conversation about them for a year or two.

AI is in that phase right now. Which makes it hard to think clearly about, because the useful parts and the hype parts are all mixed together.

Here's what I've actually observed, from using it in my own work at Reaction21 and from helping clients evaluate it.

What's Genuinely Working

Drafting and editing. AI is remarkably good at taking a rough idea — bullet points, a paragraph of notes, a vague direction — and turning it into a coherent first draft. Not a final draft. But a first draft that's good enough to edit from.

This matters because blank-page paralysis is a real productivity killer. The intimidating part of writing a proposal, or a blog post, or a policy document, isn't the writing. It's getting started. AI handles that well.

Structured repetitive work. Anything that follows a clear pattern benefits from AI automation. Code generation for repeatable patterns. Report generation from consistent data. Document creation that follows a template. These aren't glamorous use cases, but they add up to real time savings.

At Reaction21, I automated the entire pull request review process for repetitive code patterns. Every PR gets a mechanical review before I look at it. It's caught real issues. Not everything — code review still requires judgment — but the low-hanging stuff is handled.

Research and synthesis. AI is good at quickly synthesizing information from multiple sources. Not at doing original research, not at generating facts you can trust without verification, but at taking five documents and giving you a clear summary of what they say. That's useful.

Writing assistance for non-writers. This might be the most underrated use case. A lot of people in business are perfectly capable thinkers who struggle to get their thoughts into clear prose. AI helps bridge that gap. Engineers who need to write requirements. Operators who need to document processes. Managers who need to write performance reviews. They're not bad writers — they just spend their time on other things.

What's Still Hype

"AI strategy" as a thing. Most AI strategies I've seen are decks. Long decks. They describe a desired future state, outline use cases that "could" be implemented, reference industry benchmarks, and conclude with a roadmap that nobody follows.

The useful alternative is: pick one process that costs your team significant time. Automate it. Measure what you saved. Then decide if you want to do more.

The companies getting real value from AI right now are doing something boring and specific. Not transforming everything.

AI that doesn't need oversight. You will see vendors selling "fully automated" AI solutions with no human in the loop. For routine, low-stakes, high-volume tasks — fine. For anything that touches customers, creates legal or compliance exposure, or handles money — be skeptical. AI makes mistakes with complete confidence. It doesn't know when it's wrong. Build the oversight in before you find out the hard way.

General-purpose AI tools as enterprise solutions. Giving your whole team ChatGPT access and calling it an "AI program" isn't a program. It's an expense. The teams that get real value are the ones who've built specific prompts, specific workflows, and specific expectations into how the tool gets used. That takes deliberate setup. It doesn't happen automatically.

AI replacing your best people. Won't happen in the near term. Will absolutely happen for the lowest-judgment, most repetitive work. Which is mostly work nobody wants to do anyway. The real risk isn't that AI replaces skilled people — it's that it makes bad decisions faster when deployed without appropriate oversight.

A Framework That's Actually Useful

When I'm evaluating where AI might help a business, I look for work that's:

  • High volume — done frequently, not once a year
  • Rule-based — follows a consistent process with predictable inputs
  • Low-stakes on individual errors — mistakes are catchable before they cause real harm
  • Currently slow — meaning there's actual time savings available, not just speed for speed's sake

That's it. If a process meets those criteria, it's worth exploring. If it doesn't — if it's low-volume, highly variable, judgment-heavy, or high-stakes — AI probably isn't the right tool.

The mistake I see most often is trying to apply AI to the wrong category. Someone automates a process that's already fast, or deploys AI on something high-stakes without oversight, or buys a tool that requires six months of setup for a use case that happens twice a quarter.

Starting Small, on Purpose

The businesses I've seen get the most from AI have one thing in common: they started with a single, specific, measurable thing.

Not a pilot program. Not a taskforce. Not a vendor evaluation process. One process. One team. One before-and-after metric.

It sounds insufficiently ambitious. It isn't. The single-process approach is what proves value quickly enough to justify the next project. It's what generates the organizational learning that makes the second implementation go faster. It's what builds the internal credibility that gets AI into more places over time.

The "transform everything" approach usually leads to an expensive, chaotic, inconclusive implementation that makes the next AI conversation harder.

The Honest Version

I use AI in my business every day. I think it's genuinely useful and I think it's going to become more useful over time. I also think a lot of what's being sold right now is somewhere between premature and dishonest.

The useful frame isn't "how do I build an AI strategy." It's "where am I losing the most time to work that doesn't require my judgment."

Answer that honestly, pick the best candidate, and try it. See what it costs and what it saves. If the math works, do another one.

That's the whole approach. Everything else is just decoration.

If you're trying to figure out where AI and automation actually make sense for your business, an AI Workflow Audit is the fastest way to get there — 1-2 weeks, a prioritized list of opportunities, and ROI estimates for each.