Cheaper to Build, Harder to Be Heard

In 2024 I worked with 1000 Black Voices and the UK consulate to support a cohort of five BAME-led startups preparing to expand into the US. The work was pro bono: facilitating meetings with investors and customers in the SF Bay Area, and providing coaching and mentoring on what landing in that market actually involves, a subject we have written about separately in when to expand internationally.

My reason for doing it was straightforward. I am a white male with thirty years in the industry and a well-developed network on both sides of the Atlantic. Those introductions cost me a few emails and some goodwill. For the founders receiving them, the same introductions were effectively unobtainable. Lowering the ladder is cheap for the person holding it and expensive for the person who has to build their own.

Two years on, with Black History Month in the UK running under the theme of Honouring Our Communities, it is worth asking what has actually changed. Quite a lot, and almost none of it in the place you would expect.

The gap we have stopped measuring

Extend Ventures remains the only organisation to have measured this properly in the UK. Its 2023 report, covering 2013 to 2023, found that ethnically diverse founders were involved in 11% of venture rounds and captured around 9% of investment value. Black founders specifically accounted for 1.6% of rounds and 0.9% of value. Black women received 0.14% of investment, which is an improvement on 0.02% in the preceding decade and still a rounding error. Black founders’ share of value peaked at 1.13% in 2021, in the window after the Black Lives Matter protests when every fund had a statement to make, and had slipped to 0.95% by 2023. One methodological point should always travel with these numbers: Extend classifies founders by perceived ethnicity rather than self-identification, which is a defensible choice for assessing how capital allocators behave, and a limitation worth stating.

The more uncomfortable fact is what has happened since. There is no published UK figure for Black founders’ share of venture capital covering 2024, 2025 or 2026. The British Business Bank’s 2026 equity tracker analyses founder diversity by gender only. Every recent report quoting a Black founder percentage, including the Startup Coalition’s December 2025 work, is recycling the same 2023 dataset. The most-cited statistic in UK conversations about founder diversity now describes a period that ended three years ago.

The policy lever that would have forced disclosure in the wider economy has stalled too. The government confirmed in March 2026 that it would make ethnicity pay gap reporting mandatory for employers with 250 or more staff, then left the enabling Bill out of the King’s Speech in May, with no replacement timetable. Where we can see the trend, it is not encouraging: the Parker Review in March 2026 found Black directors held 2.3% of FTSE 100 directorships, down from 2.4% a year earlier, with FTSE 250 Black directorships falling from 42 to 37, and just 1.3% of FTSE 100 UK-based senior managers identified as Black.

It is difficult to argue about a gap that nobody is currently counting. That absence is worth naming during a month devoted to honest accounting.

What AI actually changed

The cost of building software has fallen, and unlike the funding picture, this part is well measured.

Jellyfish analysed around 20 million pull requests across roughly a thousand companies. Median AI tool adoption among their engineers went from about 22% in mid-2024 to around 90% by the end of 2025. Moving from no adoption to full adoption correlated with roughly double the pull request throughput and a 24% reduction in cycle time. In Supabase’s 2026 survey of around two thousand startup builders, 61% said more than half their codebase was written by AI, and 61% were solo founders, up from 53% a year earlier. Stripe’s 2026 annual letter reported that its 2025 cohort of new startups was growing about 50% faster than the 2024 cohort, and that the number of companies reaching $10m of annual recurring revenue within three months of launch had doubled year on year.

So the engineering barrier is genuinely lower. A founder who can specify clearly and review critically can now get to a working product with a fraction of the team that would have been required five years ago.

What has not happened is the thing people often assume follows. Seed rounds have not got smaller. They have got larger, with an AI premium on top, and the graduation rate from seed to Series A has worsened rather than improved. Cheaper code has not meant less capital; it has meant more capital concentrated behind fewer people. If the pitch is that AI lets under-funded founders need less funding, the data does not support it. The money still arrives with the expectations and the agenda we described in VC cash is not a training budget. What AI changes is what the money buys. Less of it goes to building the thing, and more of it has to go to the problem that is now harder.

Where the money actually went

The UK numbers show what that looks like in practice. In 2025, UK smaller businesses raised £12.3bn, down 4% on the year, but the shape of the market changed more than the total. Deal numbers fell 17%. The ten largest fundraisings took 23% of all investment, the highest concentration since 2020. Average deal size rose 17% to £6.7m while the median fell 13% to £1.0m, which is the signature of a market held up by a handful of enormous cheques. Seed was hit hardest: deal counts down 27%, median seed deal size flat at £0.6m, and the median gap between rounds stretching from 12.4 to 14.4 months. UK seed activity fell further than in the US or the rest of Europe.

In 2026 the headline reversed while the structure did not. UK AI startups raised $12.6bn in the first half, more than four times the same period in 2025 and roughly three quarters of all UK venture capital. The market is up and concentrating simultaneously.

That combination matters for anyone arguing that AI has levelled the field. A market funnelling record sums into fewer, larger, AI-heavy rounds rewards pattern recognition, prior exits and warm introductions. It is structurally hostile to first-time founders who are not already known to the people writing those cheques, whatever their product can now do with a fraction of the engineering budget.

The bottleneck moved to distribution

In that same Supabase survey, the hardest problem founders reported was customer acquisition, at 32%. Technical complexity came in at 11%, and had fallen twelve points in a single year. Fundraising was 13%. The sentence one respondent offered sums up the shift: building is the easy part, distribution is the hardest.

Then the finding that should stop anyone who cares about equitable access: 56% of these companies’ first customers came from the founder’s personal or professional network. Two thirds had never tried paid acquisition. Only 11% had built any kind of community around their product, and those who had converted users at a materially better rate.

Read those two facts together. The constraint that AI has removed is the one that money and talent could always solve. The constraint that remains is the one that is distributed by birth, schooling, employer and postcode. If first customers come from who you already know, then a founder with a thin network is structurally disadvantaged in a way that no amount of cheap inference fixes.

The enterprise end of the market tells the same story from the other side. Forrester’s 2026 survey of nearly 18,000 business buyers found 94% now using AI somewhere in their buying process, with generative AI and conversational search named as a more meaningful information source than vendor websites, product experts or salespeople. But buying groups have grown to around 13 internal stakeholders plus nine external influencers, and buyers compensate for AI’s unreliability by seeking validation from peers, analysts and experts they trust. More AI in the funnel has increased the weight of human proof, not reduced it.

That proof has a procurement dimension too. Security questionnaires now routinely carry AI-specific sections: model provenance, training data rights, hallucination controls, subprocessor transparency, alignment to ISO 42001 and the NIST AI Risk Management Framework. A founder with a credible reference and a tidy answer to those questions moves through a review in days. A founder with neither waits weeks, if the conversation survives at all.

The playbook is being rewritten, carefully

The go-to-market motions are changing at the same time, and this is where I would urge scepticism about the confident advice currently circulating.

Inbound is genuinely under pressure. Pew Research tracked nearly 69,000 Google searches and found users clicked a traditional result on 8% of visits where an AI summary appeared, against 15% where none did, and ended the session entirely on 26% of those visits. At the same time Adobe’s analysis of over a trillion visits to US retail sites found AI-sourced traffic up 393% year on year in the first quarter of 2026, and converting 42% better than other channels, having converted 38% worse only a year earlier. Both things are true: the old click is eroding, and the new referral is small but unusually high intent.

Agentic commerce is where the gap between narrative and reality is widest. The Agentic Commerce Protocol launched with Instant Checkout in ChatGPT in September 2025. By March 2026 OpenAI had withdrawn native checkout and repositioned towards discovery, after roughly a dozen Shopify merchants had gone live and adoption stayed flat. Google announced its Universal Commerce Protocol in January 2026 with Shopify, Etsy, Target and Walmart, with checkout listed as coming soon. Visa and Mastercard are still issuing country-by-country announcements of their first live agent-executed transaction. Nobody is publishing volumes, which tells you where this is. Agent-mediated discovery is real and growing quickly. Agent-executed buying is a 2027 question at the earliest in Western markets.

The same discipline applies to the emerging advice industry around optimising for AI answers. The one critical academic survey of generative engine optimisation reviewed 45 studies and found two reproducible levers, topical relevance and context position, and no demonstrated durable effect on being retrieved in the first place, let alone on conversion. Make your product and evidence retrievable and factually extractable. Treat anyone selling certainty beyond that as selling certainty.

Even pricing has defied the predictions. Outcome-based pricing, which was supposed to sweep the market, has moved from roughly 5% to 10% adoption and flattened. Hybrid models, a base subscription plus usage, went from 25% to 37%. Pragmatism won.

Who has stepped back, and who has stepped in

The institutional picture over the last two years is not a simple retreat, and it is worth being precise about who has done what.

Some programmes have gone. Google’s Black Founders Fund in Europe put around $3.9m into 46 UK startups across three cohorts, and quietly ended: the last European cohort ran in 2023, the pages now sit in Google’s alumni section, and there was never an announcement. That is a real loss of early cheques for founders who had few other routes to them.

Against that, the state has moved in, at a scale the private programmes never reached. The British Business Bank’s £500m Investor Pathways Capital package, announced in July 2025, is aimed at underrepresented fund managers rather than founders directly, on the theory that who sits on an investment committee determines who gets funded. The first cohort landed in June 2026: up to £90m committed to ten first-time microfund managers selected from 151 applications, of whom 43% are from ethnic minority backgrounds and seven are solo general partners, writing £100k to £500k cheques at pre-seed. It includes Mustard Seed Fund I, described as the first Black-led UK venture fund investing solely in consumer brands. A further £100m tranche followed in August 2026, with the next application round opening this autumn. Pathway Fund, the UK’s first Black-led endowment fund, incubated three of the ten managers in that first cohort.

There is supporting evidence for the investment committee theory. The 2026 Investing in Women Code report found that among signatory funds where more than a quarter of investment committee members were from an ethnic minority background, 33% of 2025 deals went to teams with at least one ethnic minority founder, against 24% where committees were less diverse. Those are self-selected, diversity-committed funds and the test of “at least one founder” is a loose one, so this is suggestive rather than conclusive. It still points at the same lever.

The wider corporate retreat has also been overstated in the UK, at least in intent. The CIPD surveyed over 2,000 UK employers in May 2025 and found only 2% reporting a decreased focus on equity, diversity and inclusion, while 34% had increased their efforts. The financial regulators dropped their proposed diversity rules in March 2025 and several large employers have diluted diversity targets in bonus scorecards, but I could find no UK venture firm that has closed or paused a diversity-focused fund or programme. The quiet part is the communications, not the budgets.

Where this leaves allies

Put the pieces together and the conclusion is uncomfortable for anyone who hoped technology would quietly solve this. AI has lowered the barrier that capital was good at clearing and left intact the barrier that capital was never able to clear. Proof, references, design partners, warm introductions and the benefit of the doubt are now the scarce inputs, and they are allocated by network.

Which means the most useful thing a well-connected person can offer a founder in 2026 is not advice. There is more good advice available free than anyone can consume, and a competent founder with a decent model can get a credible strategy document in an afternoon. What cannot be generated is a specific introduction to a named buyer, a reference call taken seriously, a design partner willing to be quoted, or a seat at a table. Those are units of access, and they are worth more now than they were in 2024, because everything around them has got cheaper.

There is a second implication that the venture conversation keeps missing. If the cost of reaching first revenue has genuinely fallen, then the capital that matters most at the very start is small: the £1,000 to £10,000 that the Startup Coalition identifies as too small to interest private investors and too early for Innovate UK. That is the gap where a founder without family money stops, and it is the one least discussed, because there is no fund-sized fee to be earned in filling it. The routes that do reach further than venture capital are debt: 21% of Start Up Loans and 22% of community development finance lending in 2025 went to founders from ethnically diverse backgrounds. Better than 0.9%, certainly, but debt puts the risk on the founder’s own balance sheet rather than on investors who can absorb it. We should be honest that this is progress of a particular and unequal kind.

For founders in this position, the practical implication is to treat distribution as the product development problem. Build in public, accumulate named references early, get the security and AI governance answers ready before procurement asks, and invest in the community that will supply your first fifty customers rather than waiting for paid acquisition to work.

For the rest of us, Honouring Our Communities is a well-chosen theme this year. Community is the mechanism by which access compounds: one founder’s customer introduction becomes another’s reference, becomes a third’s first enterprise logo. We can measure a funding gap, badly and infrequently. The network gap we can actually do something about, one introduction at a time.

If you have the network, lower the ladder. It still costs you almost nothing, and it is now worth considerably more.