The AI mandate: why "do something with AI" is the worst brief of 2026

Somewhere this quarter, a board member leaned forward and said the words. "We need to do something with AI." Heads nodded. A slide was added to the strategy deck. And a capable executive walked out of that room holding the worst brief of 2026.

Not because the instinct is wrong — the instinct is exactly right. Because the brief is unanswerable. "Do something" specifies activity, not outcome. And organizations are ruthlessly good at delivering precisely what was asked: activity.

What "do something" actually produces

Follow the mandate downstream and watch what happens. The executive delegates to a team. The team, lacking a target, optimizes for the only thing they can be judged on: visible motion. Six weeks later there's a chatbot on the intranet, a pilot with a vendor whose demo was impressive, and a working group with a recurring meeting.

None of it is fake. All of it is real work by smart people. And almost none of it will exist in eighteen months, because none of it was aimed at a number anyone would defend in a budget review.

Organizations don't fail at AI because the technology disappoints. They fail because nobody agreed what "working" would mean before the work began.

We've watched this cycle across 100+ projects — and the autopsy is always the same. The pilot had no owner with operational authority. The demo had no path to production. And the whole effort had no metric except "we're doing something."

The question hiding inside the mandate

Here's the move that changes everything: translate the mandate before accepting it. "Do something with AI" is a compressed, slightly anxious version of a much better question — "Where is manual work costing us hours, and what would it be worth to get them back?"

That question has answers. Findable, rankable, defensible answers. Every operation has workflows where something arrives, someone re-types it, someone chases it, and someone reports on it. Those workflows have volumes, error rates, and loaded hourly costs. Which means they have math.

The companies getting real returns from AI in 2026 aren't the ones with the boldest vision decks. They're the ones who did the unglamorous accounting first.

The one-page brief that actually works

When the mandate lands on your desk, don't build a task force. Write one page:

Four lines. Everything else — model choice, vendor choice, build vs. buy — is a downstream decision that becomes easy once these four exist, and impossible to make well while they don't.

Why the first target should be boring

The strongest resistance to this approach is aesthetic: leaders want the first AI project to be impressive. Resist that. The right first target is almost always the highest-volume manual workflow with the clearest rules — document intake, report assembly, data reconciliation. The boring ones.

Boring targets have three properties that flashy ones don't: their baseline is easy to measure, their success is impossible to dispute, and their payback funds the second project. A working invoice-intake system that returns eleven hours a week does more for your AI credibility than any demo ever will — because it shows up in the P&L, not just the town hall.

What to say in the next board meeting

So when the mandate comes — and it will — here's the reply that turns it into a plan: "Agreed. We'll pick the workflow where manual work costs us most, baseline it, and have a system in production against a hard number within a quarter. Here's the one-pager."

That sentence does something rare: it satisfies the urgency behind the mandate while refusing the vagueness inside it. The board doesn't actually want "something." They want to stop feeling like the company is falling behind. A named workflow, a number, an owner, and a production date is how that feeling ends.

Activity is easy. Outcomes are a decision. Make the decision first.