Your AI roadmap is a lie. Here's what you actually need.

Somewhere in your organization there is a beautiful AI roadmap. Quarters as columns, initiatives as bars, a confident arrow pointing toward “scaled adoption” in late 2027.

It’s a lie. Not because anyone is dishonest — because it’s fiction dressed as planning. Nobody can know today which AI capabilities, regulations, vendors and costs will hold in 18 months. In a field where the ground shifts every quarter, a three-year Gantt chart isn’t a plan. It’s anxiety management.

I say this as someone who has built roadmaps for 26 years, through every wave from the early web to AI. I believe in roadmaps. That’s exactly why I recognize the counterfeit.

What a fake roadmap does

A fake roadmap exists to make uncertainty feel governed. It gets presented twice a year, updated by the PMO, and consulted by no one making an actual decision. Its real function is emotional: it lets a steering committee feel that the future has been handled.

You can spot one with a single test: has anything ever been removed from it because reality disagreed? A roadmap that only grows is a wish list. A roadmap nobody argues about is a dead document. Real plans generate friction, because real plans force choices.

The AI versions are especially fictional, for a structural reason: they sequence outcomes (“Q3: deploy agents in customer service”) before anyone has verified the two things AI outcomes actually depend on — whether the use case is verifiable, and whether the people concerned will engage. Neither is knowable from a planning workshop.

What worked instead, every time

At IRISnet we took a different route, and it’s the reason a small team took a product from zero to company flagship in three years. We didn’t start from a three-year feature plan. The cycle was: collect the updated strategies, user feedback and stakeholder needs; rationalize the requests; run them through a prioritization matrix; estimate at high level; commit to the next slice; validate; repeat. The roadmap existed — but as a living decision instrument, re-negotiated against reality at every cycle, not a mural.

And before any of that, there was something more important than a roadmap: a direction everyone could see — one coherent product as the cornerstone of a user-centric, omnichannel architecture, open to whatever came next (at the time we said IoT and XR; then AI arrived, and the architecture welcomed it). The direction was stable for years. The plan changed constantly. That’s the correct ratio.

Later, working on future-oriented strategy at an energy utility, the lesson got a name: in an unstable environment you don’t plan one future, you prepare several — future scenarios, and roadmaps conditioned on them. Since 2020 I’ve considered linear roadmaps structurally obsolete for anything touching emerging technology.

What you actually need

Three things, none of which is a Gantt chart:

1. A direction people can see. One page. Where we’re going, why, and what we are explicitly not doing. Stable for years. If your AI direction can’t be stated without the word “leverage,” it isn’t one yet.

2. The next verifiable step. Not the next twelve steps — the next one, chosen because you can tell whether it worked. AI adoption advances fastest where output can be verified; your plan should exploit that, not ignore it. Ship something small into real workflows, measure it against a criterion agreed in advance, then decide the step after.

3. A cadence of honest re-decision. A recurring moment where the plan is confronted with what actually happened and pruned accordingly. This is the discipline that separates a decision instrument from a decoration. It’s also the part organizations skip, because pruning requires admitting.

Direction, verified step, honest cadence. That triad has carried every successful transition I’ve been part of — and its absence explains most of the AI programs currently stalled in pilot purgatory while their roadmaps claim everything is on track.

The roadmap isn’t the bridge. It’s the drawing of a bridge. People cross bridges.