To lead with AI, leaders have to use it first

Plenty of companies have already tried artificial intelligence (AI) in some form. Someone bought licenses, someone ran a lunch-and-learn, and a few people now use a chatbot every day. Then progress flattens. The tools are there, but the way the company works has not changed.

Our view is that AI adoption stalls when leaders delegate it. Leaders who use AI in their own work make better decisions about where it belongs, set norms their teams find credible, and model the behavior they ask of others. Leaders who hand it to the technology team or to one enthusiastic employee end up approving tools they cannot evaluate and writing policies nobody follows.

The research points at leaders, not employees

A McKinsey survey of 3,613 employees and 238 C-level executives, published in January 2025, concluded that the biggest barrier to scaling AI is not employees but "leaders, who are not steering fast enough." Executives estimated that 4 percent of employees use generative AI (the kind of AI that drafts text, summarizes documents and answers questions) for at least 30 percent of their daily work; the actual share was three times greater.

Boston Consulting Group (BCG), the management consultancy, reported in June 2025 that the share of employees who feel positive about generative AI rises from 15% to 55% with strong leadership support. Its June 2026 edition found that 74% of frontline employees now describe themselves as AI users, yet only a third say leadership's communications about AI are clear.

For smaller firms, the Organisation for Economic Co-operation and Development (OECD), an intergovernmental body that studies economic policy, surveyed more than 5,000 small and medium-sized enterprises (SMEs). Its November 2025 report found that a third or fewer of the SMEs using generative AI were training staff, setting internal guidelines, or researching legal and regulatory issues.

Use is rising. Direction, training and norms are lagging. Those are leadership jobs.

1. 01Personal practiceReal tasks from your own week, every day
2. 02One team workflowRedesigned, not a tool added on top
3. 03Team normsWhat data goes where, disclosure, review

Share and measurePairs, show-and-tell, time saved on chosen tasks

Start with a daily practice built from your own week

The fastest way for a leader to learn AI is to use it on work already on the calendar. Real tasks, fifteen to thirty minutes a day, for a few weeks, in a tool your company has approved.

Good starting tasks:

• Preparing a meeting. Share the agenda and the last notes, then ask what questions are likely and what you might be missing.
• Summarizing a report. Ask for a summary, then ask what the report does not say. Check both against the source.
• Drafting a hard email. A delayed delivery, a price increase, a no to a long-time client. State your intent, get a draft, then rewrite it in your own voice.
• Thinking through a decision. Describe the options and ask the model to argue against the one you prefer.

Keep a short log: the task, what worked, what you had to fix, and roughly how long it took. After a few weeks you will know, from your own work, where AI helps, where it wastes time and where it is confidently wrong. That knowledge is what makes your later decisions credible.

One example of the people-first approach

Lead with AI, founded by Daan van Rossum, is one organization built around this idea. It describes itself as a "people-centric AI transformation partner" whose aim is to "create measurable value by making work better for people." Its cohort courses teach leaders practical use in about 30 minutes a day and name their own frameworks: CODO for prompting, GED-RT for prioritizing which work to hand to AI, and OVER for redesigning a workflow. The principle holds with any provider, or none: short, daily and applied to the leader's real work.

Move from personal use to one redesigned team workflow

Once personal use is a habit, pick one team workflow. Choose something frequent, painful and low risk: turning intake calls into proposals, preparing the weekly operations report, answering the same customer questions.

Then redesign it instead of adding a tool to it. Map the steps as they happen today. Mark which need judgment, which are assembly, and which exist only because information lives in two systems. Decide where AI drafts, where a person reviews and where nothing changes. Run the new version for a month with the people who do the work, and compare it with the old one.

BCG's 2026 report puts it plainly: "invest in redesigning work end-to-end, not in more tools."

Set team norms people can actually follow

Norms work when they are short and written by someone who uses the tools. Three questions cover most of it:

• What data can go where. Which tools are approved, and which information (client records, financials, personnel matters) never goes into a tool that is not approved for it.
• When to disclose AI use. For example, always with clients when AI shaped the substance of a deliverable, and internally whenever someone asks.
• What review looks like. Who checks AI output before it leaves the building, and what they check: facts, numbers, tone and anything with legal weight.

The OECD found that 10.5% of SMEs worry about staff using generative AI in secret, and suggests that guidelines and training can help. Clear norms turn hidden use into visible use.

What changes in the leader's job

• Judgment becomes the product. When a first draft costs almost nothing, the value moves to deciding what is good enough, what is wrong and what matters.
• Questions matter more. A well-framed problem gets useful output. A vague request gets confident filler.
• Reviewing machine output becomes routine. In BCG's 2026 survey, close to half or more of respondents report spending more time reviewing and correcting AI output or directing AI. Plan for that time.
• Redesign replaces accumulation. The question shifts from "which tool should we buy" to "how should this work be done now."
• Learning time needs protection. If learning only happens after hours, it does not happen. Put it on the calendar, starting with yours.

Reskill the team without a big program

A company of 10 to 500 people rarely needs a formal academy. It needs a few habits:

• Pairs. Match someone who uses AI daily with someone who does not, on a real task, for an hour a week.
• Show-and-tell. Fifteen minutes in an existing meeting where one person walks through one task: what they asked, what came back, what they fixed.
• A shared library. One folder with the prompts and workflows that worked, who owns each one and when to use it.
• Measure what the team chooses. Let each team pick two or three real tasks and track time spent before and after. Numbers the team chose and measured itself are the ones it will trust.

Risks to manage

• Data privacy. A client file pasted into the wrong tool costs more trust than any efficiency gain earns back. Approved tools and a clear data rule come first.
• Overreliance. The OECD report cites research showing that overreliance on generative AI can reduce performance because errors are hard to detect, especially on complex tasks done by less experienced users. Review standards exist for this reason.
• Uneven adoption. Some people race ahead while others quietly opt out. Pairs and show-and-tell narrow the gap; mandates tend to widen it.
• AI theater. Announcements, license counts and pilots that never reach real work. The test: which workflow is done differently today, and what did the change save?

What this means for your business

If AI adoption in your company has stalled, the answer is probably not another tool. It starts with the people at the top using AI in their own week, learning where it helps and where it fails, then redesigning one workflow with their team under norms everyone can follow. That sequence is less impressive to announce and more likely to change how work gets done.

If you want help building that practice with your leadership team, book a meeting with us.