Why Your AI Budget Stalls: Briefing Agents Is a Management Skill

Companies are pouring money into AI and getting little back, and the gap between the two isn't a technology problem. It's a management one: say what outcome you want, hand over the context, and define what finished looks like. Only about a third of workers have had any AI training from their employer in the past half year, and what exists mostly covers prompting and stops right where the work becomes useful. Meanwhile nobody at your company has read anyone else's prompt, so nothing builds on anything. A quick test makes the point. Ask three colleagues for the prompt each used on the same kind of task.

The spending and the payoff

A survey of a thousand C-suite executives at mid-market firms, run by the National Center for the Middle Market, found those companies directing 28% of investment dollars to AI. That's the single biggest destination, ahead of IT systems and equipment. Yet a broader Conference Board panel reported that 38% of American CEOs, the highest share of any region, expect AI to hurt their company. So the money goes in first, and the people spending it half expect to regret it. Nobody has named the thing in between.

The return is thin. Among organizations in McKinsey's survey that report any profit effect at all, most put AI's share of EBIT below 5%, and that sample leans toward firms with much bigger budgets than yours. The few getting real value stand out in two ways. They redesigned how the work gets done, and their leaders visibly own that redesign. You can't buy either one. You can purchase a deck recommending both, but the doing stays with you.

Training doesn't close the gap. The Conference Board also found that just one worker in three has received employer-provided AI training in the last six months. Where it exists, it teaches AI literacy and elementary prompting, and directing agents tends to fall outside it. The syllabus ends exactly where usefulness begins.

The skill in question

Three habits cover most of it. State the outcome. Supply the context that someone absent from the meeting, human or agent, would need. Describe what done looks like. This is the oldest material in management, aimed at a new audience. What's different is that an agent follows the brief literally, so a weak brief shows up within ten minutes. That's the missing link between the budget and the result, and it explains how one person can sign off on the spend while expecting it to flop.

Day-to-day experience backs this up, and it's the part most spending plans skip entirely. The hours that change results don't go into cleverer prompts at the moment of asking. They go in beforehand: instructions that apply across a whole system and per project, written procedures for repetitive jobs, and a shared knowledge base the agents read before touching anything. That's onboarding. Like good onboarding, it gets revised as you learn what the agents keep getting wrong. It still beats starting from zero every morning.

The mental shift that raised output was to stop treating AI as a tool. Think of it as an unlimited supply of colleagues who know a little about nearly everything, then ask how you'd brief them and how the work should change. That direction holds up: better context produces better outcomes. Whether it runs backward, so that people who direct agents well also get better with human teams, hasn't been measured. Don't bet on it.

We never taught the first version either

Management itself has a long record of going untaught. In 2023, the head of the Chartered Management Institute, Ann Francke, described the usual failure as people getting promoted for technical skill while behavior and other leadership qualities were ignored. Her institute sells management qualifications, which is worth knowing, though it doesn't make her wrong.

Plenty of managers got no preparation on their first promotion. You read books and articles, copied what you liked in the best leaders around you, and resolved not to repeat what the worst ones did. The real curriculum was whoever sat nearby, and the awful bosses occasionally left a deeper mark than the great ones.

Watching and copying doesn't carry over to a chat window where the audience is a single person. Finished work gets reviewed, but the prompt behind it stays on the screen where it was typed. The only counterparty who sees it is agreeable by design. It tells you your idea is stronger than its own, then explains at length why that idea carries the weight. (Even when you tell an assistant that flattery is a giveaway, it nods warmly and carries on.) The colleague who'd have winced at a thin prompt has been swapped for something that's motivated to be pleasant and wordy.

An exercise that costs nothing

Here's a free experiment. Request the prompt that three teammates each wrote for a similar job, then line them up and read. No money or approval is involved, only three people who'll let you look. Results will vary far more than anyone predicts, and the explanation lives in the prompts themselves. That gap is the lesson, and it's visible within a single sitting; most managers have never tried it. Whoever has built up standing instructions, written procedures and accumulated context will stand out immediately. That person is your curriculum.

Put another way, given an hour to get a good answer from an agent, spend fifty minutes on the brief. Delegating to people always worked like that too; nobody wrote it down.

You can learn much of this alone. Shared practice is what makes it compound past one individual. And skip lists of the top 100 prompts for anything, since nobody opens those twice. What helps is a syllabus, not a shortcut.

There's one objection with no good answer. Prompts stay private for a reason: they're a snapshot of how someone really thinks, captured before it got tidied up. Asking three people to lay theirs on the table costs them what finished work has always protected. That's why yours goes on the table first. The biggest line in your investment budget still depends on a skill nobody has watched anyone else practice.

A prompt for your own syllabus

Paste this into a fresh conversation. It's the do-it-yourself version:

You are an experienced leadership coach. I direct AI agents, or my teams do, and I manage people. Build me a starting curriculum for both, treating them as one skill with two different counterparties.Cover the part they share first: stating the outcome, supplying the context somebody outside the room needs, breaking work into parts that can run separately, defining what done looks like, and holding a quality bar on what comes back. Then the part that applies only to people: motivation, feedback, careers, and the conversation nobody wants to have. Then the part that applies only to agents: standing instructions, reusable context, and telling the difference between a weak model and a weak brief.For each, give me the plain-language 101, the failure mode a first-timer falls into, and the named books or freely available sources you are confident exist. End with the one thing to practice this week and how I would know it is working.