The national AI strategy, translated for a small business

Since the spring of 2026 the national artificial intelligence strategy has returned to the press at every stage, and each article gives the impression that a deadline is approaching for businesses. None is.

A strategy is a document in which the state sets a direction for itself: it commits administrations, budgets and programmes. It creates no obligation for a private business, and what obliges you today comes from elsewhere — a 2018 law amended in 2025, unrelated to it.

This article says where the text stands on the day it is written, what it actually contains, through which channels it will reach you first, and what it will not settle. It rewrites no communiqué and reproduces no numeric target: the decision that concerns you is taken on what exists.

The strategy examined in May rests on three levers presented as structural: data, digital infrastructure, and human skills. That is a public description and it is what is useful to retain.

The first lever may concern you soonest, and not in the way people imagine. A data policy is about public data, its opening and its circulation between administrations. What interests you in that is what becomes available, not what becomes required.

The second is an equipment lever — computing capacity, connectivity, hosting. It changes your purchasing options in the medium term and changes nothing about your week.

The strategy is also presented as articulated around six axes, concerned mainly with scientific research and with creating a favourable environment. That structural detail is more instructive than it looks: a strategy whose centre of gravity is research and the ecosystem acts on supply — laboratories, training, startups — and not on the demand of existing businesses.

The third, skills, produces the slowest and most durable effects, and it is also the only one a business can act on alone without waiting for anything. Section 8 says how.

The second channel is that of labels, support programmes and funding. It already exists for the startup ecosystem, and a sectoral strategy generally widens its scope.

The useful thing to know is that these schemes work on files and on criteria, and that the criteria are written somewhere before they are known. A business following the relevant body rather than the press sees the calls go past; one following the press discovers them at the closing date.

There is a design trap worth flagging. Many schemes require a track record — accounts, a registration, sometimes a declared activity in a precise field — and those conditions are prepared months in advance or not at all.

We name no scheme here, because their conditions change from one round to the next and a list published in an article is wrong by its second year. Section 10 says where to look instead.

Press monitoring is the worst way to know where a text stands: it reports stages of discussion in the vocabulary of completion, and it never reports the absence of news.

The only monitoring that says anything is official publication. A text exists when it is published; before that it is a draft, however many articles discuss it.

The cheapest practice is a quarterly twenty-minute appointment: check whether a text has been published, and if so, which and on what. Three times out of four the answer is no, and that answer is information — it tells you nothing has changed for you.

One clarification about the word “published” saves a disappointment. A published text carries a number, a date and a subject, and it can be cited. An announcement, a communiqué, a statement at a conference carry none of that and can be cited to nobody. When you check, the number is what you are looking for, and its absence is a complete answer.

That rhythm is the same as the review of an automation or a corpus, and it gains from being done on the same day. An hour a quarter covers all three.

We check the state of the text on the date you ask us, and we tell you what it imposes on you — which, to date, is nothing. We prepare the register and the list of flows, which serve in either case.

We refuse to sell compliance with a strategy. A strategy is not a standard, there is nothing to comply with, and a service sold under that name invoices anxiety rather than work.

We also refuse to repeat the announced numeric targets. They are state objectives, and turned into sales arguments they become figures nobody can defend in front of you.

What you can do without us is the whole of section 11’s list. It is useful if the text is adopted tomorrow, it is useful if it is not, and that is the only property that matters when the calendar is unknown. This article describes a state of affairs at a given date and does not replace legal advice on your own situation.