When AI Is a Distraction: A Founder's Case for Not Building It

The most useful sentence I say on intro calls is this: “I don’t think you should build AI right now.”

I say it more often than you might expect. And almost every time, there’s a pause — because founders don’t usually call an AI consultant expecting to be told not to do AI.

But the goal is not to sell you a project. The goal is to help you make a good decision. And sometimes the good decision is to wait.

The three bad reasons to build AI

Not every AI initiative is driven by a genuine business problem. Here are the three reasons I see most often that aren’t good ones.

The board asked about it. Board members read the same headlines you do. At some point in the last two years, someone on your board asked “what’s our AI strategy?” And now you need to have an answer. I get it. But “the board asked” is not a product requirement. An answer to the board question might be a slide deck. It probably doesn’t require a six-figure build.

You’re pitching and it helps the story. AI-forward language in a pitch deck does move investors, especially in certain sectors. But building AI infrastructure to support a pitch narrative is extraordinarily expensive marketing. You’re trading months of engineering time, real dollars, and organizational focus for a few slides. There are cheaper ways to make the same point.

Your competitors are doing it. Maybe they are. But do you know what they’re actually building, or just what they announced? Press releases and LinkedIn posts are not evidence of a working system. I’ve talked to founders who were in a panic about a competitor’s AI launch — and when we looked at what the competitor actually shipped, it was a ChatGPT wrapper with some branded UI on top. The right response to that is not to spend six months building your own wrapper.

None of these are reasons to build. They’re reasons to have a conversation.

What “distraction” actually costs

The cost of an AI initiative that shouldn’t have happened is not just the build cost. That’s the most visible line item, but it’s not the most expensive one.

Opportunity cost. The engineering hours you spent on an AI feature that didn’t move the business are hours you didn’t spend on the core product. For an early-stage company, this is often the most expensive cost of all. Every engineering month has an opportunity cost measured in features not shipped, customer feedback not incorporated, technical debt not addressed.

Organizational debt. Building AI systems creates obligations: the system needs to be monitored, maintained, evaluated, and eventually replaced. If the business rationale for building it wasn’t strong to begin with, you’ve created ongoing overhead that serves no clear purpose. I’ve seen companies with zombie AI systems — still running in production, still consuming infrastructure cost, and nobody can remember exactly why they exist.

Credibility with your team. Engineers are smart. When they’re asked to build something that doesn’t have a clear business rationale, they know it. The internal narrative around “we’re building AI because the board asked” is corrosive. It signals that product decisions are being driven by external pressure rather than user needs, and it makes it harder to recruit and retain people who want to work on meaningful problems.

The math on “not yet”

Here’s the thing about AI that makes waiting genuinely attractive as a strategy: the technology is improving faster than most companies can execute.

A model that would have required fine-tuning twelve months ago now performs adequately out of the box. Infrastructure that required dedicated ML engineering to operate is now available as a managed service. The pricing on API access to frontier models has dropped dramatically and will keep dropping.

If you’re at the early stages of thinking about AI — before you have a clear use case, before you have the data, before you have the internal champion to own the system — then waiting six months doesn’t just save you money. It means you’ll start from a better position when you do build.

I’m not saying wait forever. I’m saying wait until you can answer these questions:

What specific, routine decision or task are we automating?
What does success look like, and how will we measure it?
Who owns this system after it’s built — who is responsible for evaluating it, monitoring it, and improving it?
Do we have enough historical data to evaluate whether it’s working?

If you can’t answer all four, you’re not ready to build. And that’s okay. Getting ready is real work. It’s just different work than building.

When “not yet” becomes “now”

There are genuine signals that tell you the timing is right.

A specific, painful operational problem. Not “AI could probably help us with X” — a specific thing that is breaking your operations right now, with a clear volume (it’s happening X times per week) and a clear cost (each instance takes Y minutes and Z dollars). When that’s the starting point, you’re solving a real problem with real metrics.

Someone internally who is bought in. Not just the founder. A person who will own the system after it’s built, who understands the problem deeply, and who will be accountable for whether it works. Without this person, the system will be built by consultants, handed off to no one, and slowly rot.

Data that already exists. The decisions being made, or the tasks being performed, are already being logged somewhere — in your CRM, your support tickets, your operations data. You’re not starting from zero. You have material to evaluate against.

A clear rollout plan that includes human review. The first version of any AI system should have humans in the loop. Not as a permanent state, but as a transition. If your rollout plan goes straight from “no AI” to “AI makes decisions autonomously,” you’re skipping the steps that let you catch and correct errors before they compound.

When all four of these are true, you’re ready to build. Before then, you’re investing in becoming ready — which is a different, but legitimate, kind of work.

The hardest part of this conversation

The hardest thing about telling a founder to wait is that it feels, from the outside, like I’m telling them to fall behind.

I don’t see it that way. The companies that win with AI are not the ones who started first. They’re the ones who started with the right problem, built with real rigor, and shipped something that their users actually depend on. A six-month head start building the wrong thing is not a competitive advantage — it’s a mortgage on your future engineering velocity.

If you’re questioning whether your AI initiative is driven by a real business problem or by external pressure, that’s worth examining honestly. The best context for that conversation is a free fifteen-minute call. I’ll ask you some direct questions and give you a direct answer about whether now is the right time.