An open letter to Anthropic

Small businesses are being left out of AI.

Dear Anthropic,

What’s the plan for small businesses?

I ask because, looking at the way the AI industry is being built right now, it doesn’t seem like there is one. Not at Anthropic, and not at the labs around you. The buildout is real, the partnerships are huge, the spending is enormous. Small businesses are nowhere in any of it.

(I’m writing to you, Anthropic not because I think you are the worst offender - I don’t - but because I think you are the most likely to do something about this.)

Your hiring page tells the story

If you want to know who a company is building for, look at who it’s hiring. Anthropic’s open roles right now include an Art Director for Enterprise, a Copy Lead for Enterprise, a Head of Customer Programs, a Customer Programs Manager for Customer Advisory Boards, GTM Strategy for Enterprise, Field Marketing for Industries, and a stack of others built around the same word.

Below enterprise sits the startup function. A Field Marketing Manager for Startups. A Startup Account Executive. A GTM Programs Lead with explicit responsibility for partnerships with VCs and accelerators. The qualifying assumption is clear: these are the businesses worth investing in early because they might one day be enterprises themselves.

Between the two, one role. Manager, Mid-Market Industries Sales. A single mid-market sales manager selling into companies “roughly 500-2500 employees”.

Nothing for small businesses. No SMB function. No SME programme. No team whose job is to make AI work for the kind of businesses that make up 99% of the UK economy, employ more than 60% of the workforce and generate 51% of the turnover.

Project Glasswing tells the same story

Earlier this month, you launched Project Glasswing. Claude Mythos Preview, a model so capable at finding software vulnerabilities that you decided not to release it publicly, has instead been given to a coalition of partners: Amazon, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Plus around forty other organisations responsible for critical infrastructure. Plus $100 million in usage credits.

The intent of Glasswing is good. Securing the world’s most critical software is important work. That isn’t the question.

The question is what it means for everyone who isn’t in the room.

The 27-year-old vulnerability in OpenBSD that Mythos Preview found gets patched in JPMorgan’s stack on day one. The same vulnerability sits in the same underlying code that a 12-person agency, or a family-run logistics business, or a UK manufacturing SME is also depending on, and they’re left waiting. Not because anyone wishes them ill. Because they’re not the partners. They’re not in the early-warning circle. They’re not the customers Anthropic builds programmes around.

Once the same class of capabilities reaches the wider model market - through other labs, through eventual public release, through bad actors - small businesses are more exposed precisely because they weren’t there at the start. Glasswing makes the largest organisations safer first. It makes the smallest organisations relatively less safe.

There may not be a clean way to fix that. But it’s a serious question, and right now nobody at Anthropic appears to be asking it on behalf of only one type of business.

Why this matters beyond Anthropic

The numbers tell the story plainly.

UK large firms with 250 or more employees are now at 44% AI adoption. UK small firms with fewer than 50 employees are at 26%. The gap nearly doubled between 2023 and 2025, and Cambridge’s Bennett School of Public Policy describes the UK as a “two-speed race” where larger firms are pulling further ahead even as the starting gun is still being fired.

The biggest barrier isn’t access to the tool. It isn’t even cost. It’s skills, time, and knowing what to do. The UK Government’s own AI Adoption Research found that 60% of businesses cite limited AI skills and expertise as the main blocker, and 71% said they hadn’t identified a clear use for AI in their organisation. None of that is solved by another enterprise pilot.

Small businesses make up 99% of the UK business population. If AI productivity gains continue to concentrate in the largest firms, the economic impact on the country is significant. But the bigger story isn’t economic. It’s that the businesses that employ most people in this country are being structurally excluded from the most consequential technology shift of the decade.

“But you can buy Claude Pro”

Someone reading this will point out that Claude Pro is £20 a month and any small business can buy it. This is true. It also misses the point.

The gap isn’t access to the tool. It’s everything around it.

Enterprises don’t buy Claude Pro. They buy Claude alongside an account team, a deployment plan, sector-specific guidance, training, advisory boards, customer programmes, and a partnership with Anthropic that includes input into how the product evolves. The tool is the cheapest part of what they’re paying for. The expensive part is the enablement layer - the human, organisational, strategic infrastructure that makes the tool useful inside a real business.

Small businesses can buy the tool. They cannot buy the enablement layer. And without it, the tool sits underused, badly used, or unused, which is exactly what the adoption data shows.

What would actually help

You don’t need a long list of demands from me. But it’s worth being concrete about what an SME-focused programme could look like, because vague critiques are easy to nod along to and ignore:

1. Training designed for time-poor leaders, not enterprise teams. Three hours, not three weeks. Built around the way people who run small businesses actually have to learn - in the gaps between meetings.

2. Deployment guidance written for a 20-person business, not a 20,000-person one. The advice an enterprise gets from McKinsey doesn’t translate. Smaller businesses need their own playbooks, written by people who understand operating without an IT department.

3. Sector-specific use cases. Marketing agencies. Accountancy practices. Independent retailers. Manufacturing SMEs. The patterns inside each sector are recognisable to operators in that sector, and almost invisible to anyone who has only ever sold to enterprise.

4. Community access. The single most useful thing a small business can have is other small businesses to compare notes with. Anthropic could convene that. It would cost very little and matter enormously.

5. Security parity. If you’re going to give early access to defensive AI capabilities to JPMorgan, build a proportional path for the rest of the economy too. Open source maintainers being included in Glasswing is a good start, but it isn’t a substitute for thinking about the businesses that depend on that software.

6. A senior person whose job it is to think about all of the above, full-time. Not a fraction of a mid-market manager’s attention. Someone whose remit is small business, the way somebody else’s remit is enterprise. And a team to interact directly with small businesses.

I’m not the only SME owner willing to help, but

When Anthropic is ready to take small businesses seriously, I’m available.

Available to get involved. Available to help you define this. Available to follow and support whoever you decide to put in charge of it. Available to provide feedback from the sidelines, alongside the many other small business leaders who care about this and would say the same.

The reason I’m offering - and the reason others would too - is that this matters. Not just for the businesses I work with, but for the economies, communities, and lives those businesses sit inside. The AI buildout is going to define the next decade. It would be a serious failure to design it in a way that excludes 99% of the businesses it’s supposed to serve.

Sincerely,

Heather Baker

Founder, The AI Edit and Humans in the Loop