Jun 12, 2026
I should be the last person writing this article.
I run a company that builds AI systems. I've trained more than 15,000 engineers in AI and modern development. My own products use LLMs in production. AI is, quite literally, how I make my living.
So take it seriously when I tell you: most products should not add AI right now. Not because AI doesn't work — because "add AI" is not a strategy. It's a feature request from your anxiety.
The pressure is real. Your board asks about your AI roadmap. A competitor announced an assistant. Your last fundraise conversation included the question. Somewhere, a churned customer mentioned ChatGPT.
So the roadmap gets an AI quarter. And here's what I've watched happen next, repeatedly: the feature ships, the launch post performs well, and 90 days later usage settles at a number nobody puts in the board deck. The constraint that was actually capping growth — the positioning nobody could repeat, the trial that never activated — is still there, untouched, one quarter older.
That's the feature tax: the cost of building AI isn't just the build. It's the constraint you didn't fix instead.
When the "AI leverage" section of one of my audits says build, it's because the answer to all three of these was yes. Most products fail the first one.
Diagnose first. If your growth is capped by unclear positioning, a leaky funnel, or churn, AI attacks none of those. (Here's the triage I run.)
The test: write down, in one sentence, the business problem the AI feature solves, with a number attached. "Support tickets cost us 40 hours a week; an assistant that deflects a third of them is worth roughly a part-time hire." That's a reason. "Customers expect AI now" is not — customers expect outcomes; AI is one way to deliver some of them.
The AI features that survive are the ones embedded in the thing users already do — drafting inside the workflow, triaging what already arrives, summarizing what they already read. The ones that die are bolted on beside the product: the chat bubble in the corner that answers questions nobody was asking there.
If you removed the feature after 90 days, would users complain? If you're not confident they would, you're building a demo, not a feature.
An LLM feature that's 90% right feels 100% broken to the user who hits the 10%. Doing it properly means evaluation, guardrails, monitoring, and a maintenance budget that continues after launch — models change, prompts rot, costs move. I build these systems daily; the build is the cheapest part. If the plan is "one sprint and a demo at the all-hands," skip it. A reliable boring feature beats an unreliable impressive one every time a renewal comes around.
To be clear about the other side: AI earns its place constantly when the three questions pass. Internal operations are usually the first honest win — the repetitive processing, routing, and drafting work inside your own business, where errors are cheap and hours are measurable. That's where I point most SMEs first: automate your own operation before you ship AI to customers. The payback is faster and the failure modes are private.
And for some products, AI genuinely is the product. If that's you, the questions above still apply — they're just easier to pass.
One more, for founders already building AI-first: "AI-powered" is not positioning. It's an ingredient, and ingredients don't differentiate for long — every competitor has access to the same models. If your homepage leads with the technology instead of the outcome, you have the same constraint as everyone else on this page: customers who can't repeat back why you win. The model is your engine. Nobody buys a car because of the engine's brand.
AI is a multiplier. Multiply a working strategy and remarkable things happen — I watch it weekly. Multiply a broken one and you get the same broken business, with higher infrastructure costs and a quarter of roadmap gone.
Find the constraint first. If AI attacks it, build it properly. If it doesn't, write that down somewhere your board can see it — that sentence is worth more than the feature.
The "AI leverage" lens is one of four in the free growth audit I run — including, explicitly, the list of AI things your business should skip. Written memo, 7 days, no pitch at the end.