The gap AI hasn’t closed

Small businesses were promised that AI would level the field. For the price of a subscription, a family shop could have what only the big chains used to afford. The Bank of America Institute still puts small-business labour productivity at half that of larger firms.

From where I sit, the gap is a management problem. I run my own team of AI agents, and the tools were the easy part. The hard part was deciding which jobs they may do alone, which they must never touch and where they hand the work back to me. Large firms pay someone to make those decisions. In most small firms, it falls to the owner, on top of everything else.

The numbers back this up once you read what each one counts. In the US, the share of small firms using AI to produce what they sell rose from 6.3% to 8.8% in six months. The Small Business Administration puts them about a year behind large firms. The OECD surveyed small firms that already sell on digital platforms, a more digital group than most. Of those, 61% use some AI. Three in four of those users are what the OECD calls novices, running off-the-shelf tools for isolated tasks. The buying is catching up. The way the tools are used is still shallow.

8.8^%
of US small firms use AI to produce what they sell, up from 6.3% six months earlier
76^%
of small firms that use AI are novices, with off-the-shelf tools for isolated tasks

Two surveys of two different groups. The US figure covers all small firms, and the OECD figure covers small firms that sell on digital platforms. Sources: US Small Business Administration, 2025, and OECD D4SME Survey, 2026.

Picture the owner of a family shop. She is the manager, the buyer, the bookkeeper and the only full-time member of staff. Half of the small firms that have yet to adopt generative AI say their people lack the skills, and nearly four in ten name lack of time for training as a barrier to going digital. The set-up would land on her too, in an afternoon she has already spent.

What she needs is the decision a large firm pays someone to make. Where does the model belong, which work stays with her and where does it hand back? Answered well, it turns twenty browser tabs into one connected loop. The loop remembers the customer, follows up without being told, flags the regular who is drifting away and drafts a reply she reads and sends in her own voice. The handwritten note and the call that should stay a call remain hers.

Adoption · buys the tool
Collects tools
Twenty tools and twenty tabs, and none of them connect.
Orchestration · decides the work
Organises one loop
Every tool has a place, the machine carries the repetition and the owner keeps the judgement.

Same tool, two outcomes. The advantage is the decision about where the model goes.

Here a small firm has the advantage. The decision costs nothing and comes from knowing what the business is for. A family shop can settle it faster than a company with layers of sign-off, because the person who knows the work is the person who decides. It is the decision I make for my own agents, and the one I design for in products where a wrong call is expensive.