Chapter 6
Skills (.md)
The Utopia of the Agentic Enterprise, Part I. The Jobs the Agents Are Here to Take
Every few years the World Economic Forum asks large employers what they expect to happen to work, and publishes the findings as the Future of Jobs Report. This is, of course, the same organisation that hosts the world elite arriving in private jets to talk about climate change, so take it with a grain of salt.
The report lists the jobs going away first and the skills to build after. Clerks of one kind or another have headed the declining list in every edition, with executive secretaries close behind.1 In the latest edition graphic designers and legal secretaries joined the list, which the report puts down to generative AI.
The headline is cheerful: more jobs created by 2030 than lost. The new jobs are mainly in frontline work, from farms to delivery vans, and the fastest-growing are in technology specialisms. The losses are in the office, and few clerks will retrain as fintech engineers. Most of the employers plan to prioritise upskilling, while two in five plan to cut staff where AI can automate tasks, so plenty of them are planning both, probably in the same spreadsheet. Underneath the headline, the message is less cheerful. Work that follows written rules needs fewer people, which employers have been saying about clerks for a decade. And the new jobs are mostly somewhere else.
The skills list changes less. Analytical thinking has been at the top for years, with resilience, flexibility and agility close behind. They’re worth having, and someone with them is probably harder to replace. What the list leaves out is that few people get the conditions to build them, that organisations respond to a scarce capability by trying to write it down, and that the thing which resists copying isn’t on it at all. Even the skills that are on it mean something narrower once output is cheap.
Analytical thinking. AI answers fast, and just as confidently when the question is bad, so the part that stays with people is asking the right question, which takes domain intuition built from years of asking increasingly complicated questions. Then there’s checking the answer. The survey puts quality control and attention to detail near the bottom of its rising skills, which would be odd if checking were about to become the well-paid work, unless employers expect to automate that too. For the routine checks they’re probably right. Resilience, flexibility and agility. It’s the textbook view of resistance turned into a job requirement: if the change hurts, the person should bend, and the change stays as it is. Given the group of people who were asked, it probably also means don’t complain and don’t expect to pick up your kid from daycare on time, and be ready to handle more stress because you will have to do all the work with less colleagues.
Leadership and social influence. Known in the corridor as political skill. Walking into a room where four departments have different priorities and walking out with a path each can live with is something models don’t do well, because acceptance depends on trust in the proposer and on whether the process felt legitimate to the people in the room. An agent can draft the compromise in seconds, of course, but each department will bring its own agent’s draft so the compromises may not be purely mathematical in nature.
Creative thinking. Expected to rise as fast as any skill outside technology, and fastest of all, according to the survey, in insurance and pensions. When competent output is abundant, selection is the constraint, and the creative part is mostly taste: seeing that a sound building is ugly, or that a grammatical paragraph is dead.
Curiosity and lifelong learning. When the tools change on a cadence measured in months, the speed at which you pick up what’s next matters more than what you already know. True, although it’s advice that helps the people who already have the time and the stretch projects to learn on, who are rarely the ones being told to learn faster.
These are the skills employers tell the survey they need. Now ask whether your organisation has ever paid for any of them as such. No salary band pays for curiosity. Political skill does get rewarded, but through promotion into management, which is a different job. What an organisation actually values is in the pay run, and the pay run mostly rewards throughput and tenure.
Taste Becomes a Rubric
An organisation that finds it depends on a scarce capability does what it does with everything scarce, and writes it down. Taste becomes a design rubric, and asking the right question becomes a prompt library on the intranet, somewhere near the last knowledge-management initiative. Both help the people who lack the capability, which is fair enough. They also let the organisation stop paying a premium to the people who have it, which may well be the part that gets the budget approved. The capabilities last longest where they can’t be turned into a procedure, which is also where they are hardest to show on a performance review.
In 1964 the sociologist Michel Crozier published a study of a French state tobacco monopoly, a factory so thoroughly governed by written rules that seniority decided nearly everything and supervisors had almost no discretion.2 One group held power anyway: the maintenance workers. The machines broke down, the rules said nothing about breakdowns, and the maintenance men were the only people who could fix them. They guarded that position. The written maintenance procedures had, of course, gone missing, and nobody was in any hurry to replace them. Management’s answer, over the years, was to try to write rules for the one thing the rules didn’t cover, and the workers’ answer was to make sure the rules never quite worked. Crozier’s conclusion was that in a rule-bound organisation, power goes to whoever controls the remaining uncertainty, and the organisation’s reflex is to remove that uncertainty by writing more rules, which creates the next one.
The agentic enterprise is arguably the newest run of that loop, with a bigger invoice. Some of what the rules never covered will be captured, and will end up in a very long prompt. What resists is whatever the organisation can’t get anyone to write down, either because it can’t be put into words, or because the person who holds it has understood what happens to them once it’s in the system.
So what resists copying is less a skill than a position: being the person who handles what the rules don’t reach, and who can therefore be blamed, and trusted to refuse. A signature means something when the person signing could have declined to sign. If they couldn’t have withheld it, it’s really a stamp, and stamps are the cheapest thing an agent produces, in whatever quantity the process asks for.
Owning the Conditions
The capabilities are unevenly distributed because few people own the conditions that produce them. Learning on the job needs stretch projects and room to fail, which an employer either provides or doesn’t, and in a year of cost cutting it probably doesn’t. Political skill grows inside specific rooms, and who gets into those rooms depends on who already holds power there. Once developed, the capabilities are portable. The conditions that develop them stay with the employers and families that own them.
So any response to the repricing of knowledge work can’t stop at an e-learning licence and an hour a week of self-development that ends up going to meetings.
Checklist
- Which capability are you building now, with evidence: asking the right question, political skill, taste?
- Who controls the stretch projects, the rooms and the exposure those capabilities need, and do you have access to them?