TL;DR: Mid-market companies now put more investment into AI than into anything else, and most organizations credit it with under 5% of EBIT. What sits between the spend and the result is an ordinary management skill: stating the outcome, supplying the context, defining what done looks like. Only one in three workers has received any employer-provided AI training in the past six months. The existing training covers prompting and stops where the work gets useful. Nobody in your company has read anybody else's prompt, so nothing compounds. Ask three people for the prompt each used on one task.
Want to listen to this article?
Subscribers to the Orion Playbook newsletter can listen to the AI-Generated Audio version of this article for free. Already a subscriber? Log in here
Mid-market companies are allocating 28% of their investment dollars to AI, the largest single destination in a National Center for the Middle Market survey of a thousand C-suite executives, ahead of IT systems and equipment. The Conference Board asked a wider panel and found 38% of American CEOs, the highest share anywhere, expecting AI to harm their company. The money goes in ahead of everything else, and the people spending it expect to regret it. Nobody has named what sits in between.
Almost nobody is getting anything material out of that spend. Of the organizations McKinsey surveyed that report any profit impact at all, most credit AI with less than 5% of EBIT, on a sample leaning toward companies with budgets you do not have. The few that do get value differ in two ways: they redesigned the work, and their leaders visibly own that redesign. Neither is a purchase. You can buy the deck that recommends both; the doing is still yours.
The Conference Board also found only one in three workers has had any employer AI training in the past six months. Where training exists, it teaches AI literacy and basic prompting. Directing agents, that report says, tends not to be what it covers. The syllabus stops exactly where the work gets useful.
State the outcome. Supply the context a person or an agent who was not in the meeting would need. Say what done looks like. Each of those is the oldest part of management, pointed at a new audience. What changed is that an agent does exactly what the brief says, and the gap shows within 10 minutes. That is what sits between the money and the result, and why one person can fund the spend and expect it to fail.
I do this process myself every day. The time that produces the difference does not go into better prompts at the moment of asking. It goes up front: system-wide and per-project instructions, written procedures for repetitive work, and a shared knowledge base my agents read before touching anything. That is onboarding, and like any effective onboarding, it gets revised as I learn what they keep getting wrong. It still beats starting over every morning.
The framing that moved my output was to treat AI not as a tool but as an unlimited number of colleagues with deep knowledge of almost everything, then ask how I would brief them and rethink how things get done. That direction works: the better the context, the better the outcome. Whether it runs the other way, whether directing agents well makes anyone better with people, nobody has measured. I would not bet on it.
We have a track record of not teaching the first version of this job. In 2023, the Chartered Management Institute's chief executive, Ann Francke, named the failure as "promotions based on technical competence that ignore behaviour and other key leadership traits." Her institute sells management qualifications, which is worth knowing and does not make her wrong.
I got no training on my first promotion into management. I read books and articles, copied what I liked from the best leaders around me, and decided not to repeat what the worst ones did. The curriculum was the people I worked next to, and the bad ones sometimes taught more than the good ones.
Learning by watching does not survive a chat window with an audience of one. Nobody in your company has read anybody else's prompt. Finished work gets reviewed; the prompt that produced it never leaves the screen on which it was typed.
The one counterparty who does see it is agreeable. It tells you the idea you brought is stronger than its own, then explains at length why it is load-bearing. (I have told the one that helps me iterate on this article that flattery is a tell. It agreed, warmly.) The colleague who would have winced at a thin prompt has been replaced by something with every incentive to be pleasant and use long prose.
So ask three people on your team for the prompt each used on one comparable task, and read them side by side. It needs no budget and no approval. It needs three people willing to show you. The outputs will differ more than anyone expects, and the reason will be in the prompts. Whoever invested in standing instructions, written procedures, and accumulated context will be obvious. That person is your curriculum.
To paraphrase, if I had an hour to get a good answer out of my AI agents, I would spend 50 minutes on the prompt. Delegating to a person always worked that way too, and it was never written down.
You can still learn most of this alone. Shared practice is what makes it compound beyond one person. If you would prefer not to do it alone, send me a message. I work with a handful of companies on this, and I will never send you the top 100 prompts for anything; nobody opens those twice. The one below asks for a syllabus, not a shortcut.
One objection, and I have no answer to it. Prompts stay private for a reason. A prompt is a picture of how someone actually thinks, captured before it was tidied. Asking three people to put theirs on a table costs them the thing finished work has always protected. That is why your prompt goes on the table first. The largest line in your investment budget still turns on a skill nobody has watched anyone else practice.
All the images were generated with AI (ChatGPT Images, Gemini Nano Banana, Claude Opus) by Gérard Métrailler.
Food for your AI
Paste this into a fresh conversation. It is 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.
Sources
Chartered Management Institute with YouGov. "New study: Bad managers and toxic work culture causing one in three staff to walk." Press release, October 16, 2023. https://www.managers.org.uk/about-cmi/media-centre/press-releases/bad-managers-and-toxic-work-culture-causing-one-in-three-staff-to-walk/ Accessed 2026-09-01
National Center for the Middle Market. Middle Market Indicator, Year-End 2025: Revenue Growth and Investment Rebound. Fisher College of Business, The Ohio State University, December 2025. https://www.middlemarketcenter.org/wp-content/uploads/2026/08/MMI-2025-Year-End.pdf Accessed 2026-09-01
Tinkoff, Dan, Lieven Van der Veken, and Michael Chui. "The state of AI in 2026: On the road to ROI." McKinsey and Company, August 25, 2026. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai Accessed 2026-09-01
The Conference Board. "Survey: CEOs Start 2026 on Edge, Citing Uncertainty as Top Threat." C-Suite Outlook 2026, January 15, 2026. https://www.conference-board.org/topics/c-suite-outlook/press/c-suite-outlook-2026 Accessed 2026-09-01
The Conference Board. "Report: Most Organizations Are Preparing Workers for Today's AI, Not Tomorrow's Jobs." Skilling for AI: Critical Factors for Navigating AI Disruption, June 8, 2026. https://www.conference-board.org/press/ai-skilling Accessed 2026-09-01
