Experience Is Subtraction: What 35 Years in Tech Taught Me About Where AI Fits

Ask an eager team where AI fits in their business and you get a list of additions. A chatbot here, a copilot there, an automation bolted onto every process that moves. Ask someone who has spent 35 years building software and security systems the same question and you usually get something less exciting and far more valuable: a list of things to remove first.

That is the uncomfortable truth I keep relearning as an AI advisor. Experience is subtraction. The judgment clients actually pay for is rarely knowing what to add. It is knowing what to take out so that whatever remains is worth automating at all.

Systems do not die from missing features

In three and a half decades of shipping software, I have almost never watched a system fail because it lacked a capability. Systems fail because they carry too much. Every component is something that can drift, break, or quietly get expensive. Every integration is a promise someone has to keep at 2am. Complexity behaves like a loan: it feels free the day you take it on, and then you pay interest on it every month it sits in production.

AI raises the interest rate. A single AI feature brings a model, prompts, evaluation, fallbacks, monitoring, and a vendor relationship along with it. None of that shows up in the demo. All of it shows up in the operating budget. This is why the question "where does AI fit?" cannot be answered by listing what AI can do. It has to be answered by asking what your business can afford to carry.

The fix that made our AI work was a constraint

Years ago I built an AI system for sports video at a company called Egoli. We used facial recognition to find marketable moments in football matches. Pointed at every player in the Premier League, the system produced garbage. At one point it confidently tagged a player who had been dead for 20 years.

The fix was not more AI. It was less scope. We constrained the system to the 22 players actually on the pitch for that match, and accuracy soared. Nothing about the model changed. What changed was the size of the problem we asked it to solve. That lesson has held through every AI project since: the winning move is usually a constraint, a removal, a narrower question. Subtraction is not the compromise. Subtraction is the strategy.

The subtraction pass

Inside the three companies where I act as CTO, and with every advisory client, the first pass on any AI idea now looks the same. Before a model is chosen or a line of code is written, we do three things.

First, delete steps. Most workflows accumulate stages that exist for historical reasons. Automating a step nobody needs is the most expensive way to keep it. Second, collapse handoffs. Every handoff is a queue, and queues are where time goes to die. Often the "AI opportunity" evaporates once two handoffs become one. Third, shrink the scope until the value is obvious. If you still have to sell people on the payoff, the scope is too big. Cut until nobody in the room needs convincing, then point AI at exactly that and nothing else.

Half of my advisory work amounts to talking clients out of AI they do not need. That is not skepticism. I was employee #2 at an AI startup, where I serve as Senior Director of Technology, and I build with these tools every week. It is precisely because I build with them that I want them carrying the smallest possible load, in the narrowest possible lane, where the return is undeniable.

What this means for the help you hire

This is also the real difference between kinds of outside help. A vendor is structurally biased toward addition, because addition is what they sell. A fractional CTO or independent AI consultant should be biased the other way. The value of an experienced AI advisor is not enthusiasm, and it is not access to tools you could rent yourself this afternoon. It is a trained eye for carrying cost, and the willingness to say "take that out" before saying "add this."

So if you are staring at a list of places AI might fit in your business, try inverting the exercise. Write down what you would remove from each process first. The processes where that list is long are usually not AI opportunities yet. The one where the process is already lean, the mistake is cheap, and the volume is real: that is where AI fits.

The teams that win with AI are not the ones that bought the most of it. They are the ones that needed the least of it to move a number that matters.

If you want a second set of eyes on your AI plans, and the subtraction pass that goes with it, I take a few clients at a time. Book your free 15-min AI Fit Call →