AI is Great for Small Businesses
An argument for why AI is asymmetrically positive for small businesses, rewarding firms built on trust, judgment, local knowledge and owned context while punishing those built o...
AI is asymmetrically positive for small firms.
This week, Cloudflare CEO Matthew Prince warned that the rise of autonomous AI agents will “destroy small businesses” by making them invisible to consumers.
Prince argued that instead of humans searching and clicking on various websites, AI agents will increasingly make purchasing decisions on our behalf, which heavily favours established, large brands.
He thinks consumers will trust AI to find information and buy products, cutting off direct web traffic that traditionally helped users discover smaller businesses, and that this risks creating a massively consolidated market dominated by only a handful of powerful tech, retail, and real estate mega-corporations.
— (@)
I think Prince is broadly incorrect.
AI will create more small businesses than any technology in history. A single founder can now do the work of an entire team. The cost of building, operating and scaling a global company is collapsing.
Some businesses will almost certainly disappear. They always do. Millions more will be created because AI is the greatest entrepreneurship unlock we’ve ever seen.
I do think Prince is on to something though - agents do annihilate friction-rent - the income derived from the friction in finding an alternative product or service.
Many small businesses make their money from being the closer, more convenient option. But agents set search and comparison costs to zero, so a business whose only edge was being the path of least resistance is likely going to struggle.
The problem is Prince says "small business" but then describes, exclusively, a commodity reseller competing on price and proximity.
The small business that benefits from AI is the financial advisor doing home visits, the accounting partner that's known you for ten years, and the specialist whose value is judgment, taste and local knowledge.
That business is not exposed to agents in the same way.
Agentic commerce makes sense where making a choice doesn't require your taste or judgment. Selecting a flight on a given date between two airports is a decision you'd happily delegate. These are spec-matching problems rather than preference problems, and agents excel at them.
The moment you introduce taste, the agent needs a mental model of your taste, and that model has to come from somewhere.
Either you build it through repeated interaction with the agent, in which case the agent becomes the relationship and the original merchant is cut out. Or the agent falls back on generic proxies for taste, things like reviews, ratings, and popularity and at that point it isn't optimising for you. It's optimising for the median buyer.
Mega-corporations win in Prince's scenario because they’re legible to a machine. They have consistent SKUs, predictable quality, structured data, reviews at scale, a presence in every dataset an agent would be trained on or likely to query.
A lot of these SME doomer narratives assume small businesses will keep running traditional playbooks. They assume small businesses won’t adapt, innovate, or benefit from AI.
That’s shortsighted and it underestimates entrepreneurs and the resilience of small businesses.
AI is asymmetrically positive for small firms.
It helps them more than it helps the incumbents they compete against, to the point where I think it's one of the rare technologies that's regressive in favour of the small.
Many of an SME's historical weaknesses were access problems - it couldn't afford to hire, couldn't justify enterprise software, couldn't run an analyst team, couldn't build a tool shaped to its own workflow.
On the other hand, those access advantages were an enterprise’s strengths. AI doesn't gift the enterprise new capability so much as it dissolves the moats that that capability used to buy.
When the cost to summon an analyst trends toward zero, the firm whose edge was a hundred analysts on the payroll can no longer bank on that.
AI can't replace trust, judgment, or your unique knowledge and understanding of a client’s circumstances. That's exactly what the SME was already great at.
For a small growing firm, AI is labour-augmenting and headcount-increasing at the same time.
When each person, armed with AI, can run more of your decision-loops - you don't want fewer people, you want more of them, because the binding constraint has moved to judgment and attention itself and each unit of it now commands more leverage.
I think there are three conditions that determine whether this upside actually accrues to the SME.
The first is the operator.
AI is net positive for the curious SME and punishing for the incurious one. Nothing forces an owner who'd rather not adopt AI to adopt it.
The second is sovereignty.
The capability the small firm now runs on is rented from somewhere. Mostly that means OpenAI, Anthropic, or Google directly, or one of the cloud platforms reselling their models.
So the SME has to ask whether it has gained independence or simply moved its dependency upstream. If the client memory, workflows, decision logic and business data now live inside someone else’s AI platform, the firm has not become more sovereign. It has become more dependent on the company best positioned to abstract it away.
The third is what you do with the freed capacity.
AI doesn't just give your firm more capacity, it gives every firm in your category more capacity, at the same time, with no increase in the number of buyers. The result isn't more business for everyone, it's more supply chasing the same demand, and the value of each unit of output falls.
The winning firm takes its AI-freed capacity and spends it on the human layer - the home visit, the better prepared meeting, the faster follow-up, the extra judgement, the relationship work that does not commoditise.
That’s how AI becomes a moat for SMEs instead of a margin race.
But don’t take my word for it. Prince is the same man who, two months ago, laid off a fifth of Cloudflare employees while growing thirty-plus percent and hiring builders and sellers aggressively.
His stated reasoning, borrowed from Peter Drucker, was that AI automates the "measurers" - ops, compliance, finance, middle management - while the builders and sellers are safe, and salespeople specifically are safe because, in his own words, humans still control budgets and prefer to buy from humans who can build trust.
Prince's entire operating model of his own company is the trust-and-relationship layer is AI-safe; the commodity-measurement layer is what gets automated.
That’s my thesis. The small businesses worth defending are "sellers" in his own framework, they are the trust-building, relationship-holding, human-judgment layer that he just told his own staff is safe from AI.
Prince believes the relationship layer is defensible enough to bet his payroll on it inside Cloudflare.
We got into this and more this week in the Big Episode 63 of The Good Stuff.