The internet does not have a shortage of AI content. It has a shortage of honest AI content: the kind written by someone who is actually building with it inside a real business, not synthesizing a trend piece from the outside. AI adoption among small business owners nearly doubled between 2023 and 2024, and the Federal Reserve is now tracking AI adoption as a macro economic indicator. Everyone has an opinion. Almost nobody is publishing what actually works day to day.
I came to AI through research, not through the hype cycle. I spent time as an embedded advisor at an AI drug design company, watching how the technology actually functioned in a regulated science context: where it accelerated real work, where it failed in ways that mattered, and where it quietly became indispensable in ways nobody had predicted. That experience shaped how I think about adopting any tool: you test it, you track what actually happens, and stay skeptical of your own enthusiasm until the results justify it.
Now I run SOM Aesthetics, a concierge dermatology and medical spa in Encinitas, and I consult with founders and operators across industries. Having a PhD in biochemistry means I've spent my career designing experiments, controlling for variables, and refusing to accept a result until I've tried to break it. That turns out to be a useful orientation for working with AI. You don't trust the output. You test it, adjust it, and keep going until you understand what you're actually dealing with.
Here's what that looks like in practice.
First: The Mindset That Makes AI Actually Work
Most people approach AI like they approach a new hire: give it a task, evaluate the output, decide if it's good enough. That gets you mediocre results.
The mindset that actually works is closer to how I run experiments: you define what you're trying to learn, you design the test, you monitor the output, and you iterate based on what you observe, not what you expected. AI is a tool that rewards people who stay alert and are genuinely open to being surprised by what works.
The founders who use AI best aren't the ones who know what it can do. They're the ones who keep running experiments until they find out.
That framing matters because it changes how you engage. You're not looking for a perfect answer on the first try. You're running a protocol, monitoring the output, and adjusting. Sound familiar? It should. It's the same skill set that runs a lab.
Where I Actually Use AI
Agentic marketing systems
The highest-leverage thing I've built with AI isn't individual outputs. It's systems. What I'd call agentic marketing: encoding your brand: the voice, the guidelines, the positioning, the rules, so that AI produces consistent work at scale without you rebuilding the context every time.
For SOM Aesthetics, that means automated competitive research, landing page development with brand guidelines already baked in, hiring job descriptions that actually sound like us rather than a generic job board, and email campaigns built from a repeatable template system. The difference between using AI as a drafting tool and using it as a system is scale. A system produces compounding leverage. A drafting tool just saves you time on the thing you're writing right now.
Building operational tools
This is the use case that surprises people most: I've built multiple web applications for SOM Aesthetics without hiring a software engineer.
One manages clinical trial patient scheduling across multiple visit schedules dictated by an IRB-approved research protocol. Coordinating that manually, matching patients to visit windows and tracking compliance with protocol timing requirements, consumed significant administrative time. The tool handles it. I also built an AI-powered patient consultation tool built into our clinical experience. Neither required a developer. They required me to understand the operational problem clearly enough to build the solution with AI's help.
This is a different kind of leverage than most founders are thinking about when they hear "AI." If you know your operational problem well, you can build the tool that solves it. Most founders don't realize that's available to them.
Data analysis
AI is genuinely good at querying and cleaning large data sets in ways that used to take days. Recently I was working through a data analysis where patterns surfaced faster than any analyst could have found them manually. Some of those patterns were useful. But the directionality (what questions to ask, which findings actually mattered, what to do with the results) was still mine to figure out.
That nuance is worth naming explicitly. AI accelerates your ability to work through data, but it doesn't know what you're trying to learn. It can surface patterns; it can't tell you which ones are signal and which are noise. You still have to be the one asking the right questions. What changes is how fast you can answer them.
Drafting and communication
The volume of written communication in any business is enormous: client emails, internal docs, treatment protocols, marketing copy, proposals. I use AI as a first-draft engine for much of it. Not to publish unchanged, but to get past the blank page faster. When the brand voice is already built into the system, the editing is refinement: accuracy, specificity, final judgment. Not a voice overhaul.
For anything requiring regulatory accuracy, patient communications at SOM, anything touching medical claims, I'm the final filter every time. AI doesn't know what you can and can't say in a regulated environment. That's still a human job. But it dramatically reduces the time between "I need to write this" and "I have something to work with."
Research and intelligence
I spent years building what I called "get smart decks" for pharma and biotech clients: rapid intelligence briefs covering a company's competitive landscape, clinical pipeline, and positioning. It used to take days. Now I can assemble a foundational version in a fraction of the time. The output still needs expert interpretation. AI will hallucinate facts and miss nuance. The skill isn't trusting the output. It's knowing what questions to ask and how to verify the answers.
What Broke (So You Don't Have To Learn It the Hard Way)
Trusting it without verifying. The first few times I used AI for research synthesis, I caught hallucinations that were plausible enough to be dangerous. A drug name that didn't exist. A study cited with the wrong conclusion. These aren't edge cases. They're features of how large language models work. Treat every factual output as a starting point for verification, not a finished product.
Using it to replace thinking instead of accelerate it. The fastest way to make AI useless is to outsource your judgment to it. It can draft an email, but it doesn't know your client relationship. It can suggest a strategy, but it doesn't know your constraints. Use it to move faster inside your own thinking, not to skip the thinking.
Generic prompts getting generic outputs. The quality of what you get out is directly proportional to the specificity of what you put in. I spend more time on my prompts than most people do, and it shows in the outputs. Context matters. Tone matters. Constraints matter. Tell it who you are, who you're writing for, what you've already ruled out, and what you actually need.
What Surprised Me
The thing I didn't expect was how useful AI would be as a thinking partner at odd hours. Building and running businesses means the work doesn't stop when your team does. Having something you can think out loud with at 10pm that will push back on your assumptions, ask clarifying questions, and help you work through a problem, is more valuable than I initially gave it credit for.
I also didn't expect how much it would surface the gaps in my own thinking. When you have to explain something clearly enough for an AI to execute on it, you quickly discover what you haven't actually figured out yet. It's a clarifying mirror in a way that's uncomfortable and useful in equal measure.
The Bottom Line
AI is not magic. It's not a replacement for expertise, judgment, or taste. It's a leverage tool, and like any tool, it works in proportion to how well you understand what it's actually good for.
The founders who are getting the most out of it right now aren't the ones reading the most articles about it. They're the ones running experiments, monitoring what works, staying honest about what doesn't, and building systems that let them move faster without moving sloppier.
That's the only version of AI adoption worth doing. Work smarter, not just faster.