Don't Let Them Leave You Behind

Meanwhile, look at the firms you'll be compared against. McKinsey is the instructive case, and not for the reason people assume. By early 2026 the firm's own managing partner was describing a workforce of 60,000 that included around 25,000 AI agents, with the expectation that agents would soon match human staff in number and every employee would be "enabled" by at least one. The lazy reading is "robots replacing consultants." The actual story is more useful to you. McKinsey did trim some back-office roles, but it has kept hiring and training the people who sit in front of clients, solve problems, and exercise judgment. The firm decided its scarce resource was capable humans who could direct these tools, so it invested in making its people more productive rather than simply cutting headcount. That is what a commercially driven organization does when it gets a tool this powerful: it arms its best people and raises the bar.

Your organization probably will not do that, at least not on its own and not soon, and that is the trap.

Here is how it actually plays out, because it is worth being concrete about the mechanism. The commercial firms train their people now, while times are decent, because productivity is their product. They have every incentive to make a mid-career analyst twice as effective. Public-sector and quasi-public bodies feel that incentive far more weakly. There is no quarterly P&L screaming at a trade commissioner to halve the time it takes to produce a market brief. So the training does not come, or it arrives as a single optional webinar nobody is required to finish. People keep working the old way because the old way is still tolerated.

Then the cycle turns. Governments across the developed world are carrying heavy debt loads and facing real pressure to look efficient. When the cutting starts, and in most jurisdictions it eventually does, the questions asked are blunt. What does this office actually produce? Could fewer people do it? The professionals who get protected, or who land softly somewhere else, are the ones who can demonstrate they are worth more than their job title. The ones who cannot are exposed twice over. They were never trained up, so they have no edge inside the organization, and they have no marketable skill outside it either. They spent years becoming fluent in one institution's internal processes and learned little that transfers.

That is the quiet danger of a slow employer: not that it fires you in a burst of AI enthusiasm, but that it lets you stand still, comfortably, until the day standing still becomes a liability. By then the market has moved on without you. The early-career numbers already show the squeeze beginning. Researchers tracking AI-exposed fields found that early-career workers saw a roughly 16 percent relative decline in employment while roles for more experienced workers stayed stable, and McKinsey's own survey found 51 percent of organizations reporting that generative AI was reducing their need for entry-level roles. If you are young or mid-career in this profession, the window to become demonstrably more valuable than an entry-level cost line is open now. It does not stay open indefinitely.

So the message is simple and a little uncomfortable: do not wait for your organization to save you. Push it, and if it will not move, train yourself.

Pushing it means being the person who shows up with a working example rather than a complaint. Most agencies do not adopt because no one inside has made the abstract concrete. Build the market scan that used to take a week in an afternoon, bring it to your director, and show the before and after. Volunteer for the pilot nobody else wants. Ask, in writing, for the training budget, because a documented request is harder to ignore than a hallway grumble and it puts you on record as the one who saw it coming. You are not trying to win an argument about technology; you are trying to make it embarrassing for the organization to keep doing things the slow way.

Training yourself means not waiting for permission. The skills that matter in commercial diplomacy are learnable on your own time, and they are not the ones the hype cycle obsesses over. A few that are worth real investment:

Learn to direct AI through a genuine piece of your own work, not a toy exercise. Take a real country brief, a real investor profile, a real mission agenda, and learn to get a strong first draft out of these tools and then tear it apart for what it got wrong. The skill is not prompting; it is knowing what a good output looks like and where the model is bluffing, which only comes from doing it on work you actually understand.

Get comfortable with data you would previously have sent to someone else: trade flows, investment statistics, sector trends, company financials. You no longer need to wait three weeks for an analyst to pull and chart it. The professional who can interrogate a dataset and tell a minister what it means has a real edge over the one who can only commission the work.

Learn to build small, useful things. These are not enterprise systems, just the lightweight tools that make a team faster: a way to monitor a market for relevant news, a system to keep investor leads from falling through the cracks, a repeatable way to turn messy meeting notes into briefings. The barrier to building these has collapsed, and the person who builds them becomes visibly indispensable.

Develop the judgment to know what should never go near an external tool, and be the person in the room who can say so credibly: confidential commercial information, government advice, anything covered by your obligations. Fluency includes knowing the limits, and that fluency is itself scarce and valuable in a sector rightly nervous about exactly this.

You do not have to go hunting for where to begin. Some of the most credible training is free. Anthropic runs a free academy that walks you through working with its Claude models and hands you a certificate at the end. Google's AI Essentials covers the fundamentals, including prompting and the responsible-use questions our field cares about, in roughly ten hours. Microsoft offers self-paced modules through Microsoft Learn for the Copilot most public-sector staff already have on their desks. An evening a week with any of these will put you ahead of most of your colleagues, and well ahead of where your organization is likely to take you on its own timeline.

None of that requires your employer's budget or blessing. It requires your evenings for a while, and the decision to take your own trajectory seriously.

Before you decide whether to push your employer or quietly prepare your own exit, it helps to know which kind of organization you're actually in. The signals are not hard to read once you know what to look for. Run your workplace through these ten questions. They're built from what the organizations getting this right are actually doing, and the answers will tell you more than any all-staff memo about "embracing innovation."

1. 1

   Has it gone beyond the one tool it was handed?

   Many organizations switched on Microsoft Copilot because it came bundled with software they already paid for, and then stopped. That is where adoption starts, not where it ends. A serious organization is actively evaluating purpose-built tools for its real work rather than treating a single default assistant as the whole of its AI policy.

2. 2

   Does it have a clear, usable policy on what data can go into which tool?

   This is the one that matters most in our field. There should be an unambiguous, written distinction between classified material, confidential commercial information, personal data, and open information, with specific guidance on which tools are approved for each. If the only policy is a vague "be careful" email, or worse, a blanket ban everyone quietly ignores, the organization has neither protection nor permission. Both failures are dangerous.

3. 3

   Has it actually trained people, or just given them access?

   Handing someone a license is not the same as training them. The agencies pulling ahead run real programs that teach people to use these tools on their actual work. Recall that in the UK study, three-quarters of public servants who received training found the tools easy to use. The organizations that skip training aren't avoiding risk; they're guaranteeing their people stay slow.

4. 4

   Is anyone measuring whether it's actually saving time or improving quality?

   Mature organizations track outcomes, not license counts. "We bought 500 seats" is a procurement statistic; the evidence that matters sounds more like "market briefs now take two days instead of two weeks and the quality went up." If no one can point to a concrete before-and-after, the adoption is cosmetic.

5. 5

   Is the push coming from leadership, or only from a few enthusiasts?

   The organizations that succeed are driven from the top with a clear strategy. The ones that stall leave it to a handful of self-taught keeners who use these tools despite the institution rather than because of it. If everything good is happening unofficially, the institution is failing its people.

6. 6

   Is it rethinking how work gets done, or just speeding up the old steps?

   The real gains come from redesigning a workflow, not from doing the same nine-step process slightly faster. An organization asking "how do we do what we already do, but quicker" is thinking too small. The better question is whether the process should exist at all in its current form.

7. 7

   Does it know where "shadow" AI use is happening, and is it bringing that into the light?

   In a lot of agencies, people are already quietly pasting work into consumer tools because no approved option exists. A forward-moving organization knows this is happening and responds by providing safe, sanctioned alternatives. One that's falling behind either pretends it isn't happening or punishes the people doing it, which only drives the risk further underground.

8. 8

   Is there a named person or group actually accountable for this?

   Somewhere there should be a person, a team, or a council that owns AI adoption and governance and meets regularly. If you cannot name who is responsible for this at your organization, that is your answer: nobody is.

9. 9

   Is it investing in the people who could become genuinely advanced users?

   A small share of any workforce generates most of the value from these tools. The organizations that get it are identifying those people and investing in them deliberately. The ones falling behind treat everyone identically and develop no one, which means they never build the internal expertise to go further.

10. 10

    When it talks about AI, is it about doing more, or only about cutting?

    Listen to the framing from leadership. An organization that talks about AI as a way to extend its people, take on more files, and raise the standard of its work is one you want to stay in. An organization that only ever frames AI as a way to reduce headcount is telling you exactly how it sees you, and exactly when you should start preparing.

Tally it up honestly. If your organization is failing most of these, that is not a reason to despair, but it is information you should act on. It means the institution is unlikely to develop you, which makes the case for developing yourself more urgent, not less. And it means that if you become the person who can credibly answer these questions for your employer, who can stand up the data policy, run the first real training, or build the proof-of-concept that makes the savings undeniable, you become extraordinarily valuable.

There is a comfortable story going around that public-sector professionals are insulated from all this, that the pace will be gentle, the protections will hold, and there is time. Maybe. But comfort is exactly the condition that lets a slow organization keep you still until you are easy to cut.

Your organization may well be a laggard. That is not an excuse to be one yourself; it is the reason you cannot afford to be.