Distributed channels, centralised value: AI and the opportunity for automotive marketplaces

A few weeks ago I gave a talk at AutosBuzz 2026 about something I think every automotive marketplace is going to have to reckon with in the next few years: what AI and large language models are doing to the way people buy cars — and where that leaves the marketplaces sitting in the middle.

I did it slightly differently. The talk and a working AI product were built in parallel — so by the end of the session, the audience could see a real, functioning car-search agent that I'd put together with an LLM, live. The point wasn't the demo. The point was what the demo makes obvious.

Here's the argument, and where you can watch the whole thing.

The talk in one line

AI is still in its infancy, but it is going to change marketplaces — and the people who win won't be the ones who fight the disruption. They'll be the ones who absorb it.

I built the talk in three parts: the disruption, the data, and the precedent.

Part one — the disruption

To set the scene: in 2016, the natural-language car-search startup Carsnip cost roughly £3.89m and a team of thirty to build, over several years. In 2026 I rebuilt that same product layer in an afternoon — one person and an LLM.

That's not a point about Carsnip. It's a marker of how far the cost of building has collapsed in ten years. When the product layer gets that cheap to build, the moat was never the product. It's the data and the relationships underneath it.

Part two — the data

The numbers are the part people underestimate. A few that stuck with the room:

4% of all public GitHub commits are now written by Claude Code alone — and that doubled in a single month. Real AI-assisted code is closer to 30–40%. (SemiAnalysis, Feb 2026)
12× growth in AI-driven referral traffic to retail sites in seven months, while organic search fell 9%. AI is now the fastest-growing referral source, full stop. (Adobe Digital Insights, 2026)
40 million AI-influenced car-buying journeys per year by 2030 — and that's the US figure alone. (BCG / OpenAI, 2026)
The average UK buyer spends 48 days deciding, across four different car-search sites. You own one stop on a seven-week journey. (Motors Digital Touchpoints, 2025)

And the leakage isn't to a competing marketplace. It's to everywhere else at once — ChatGPT, Claude, Gemini, Perplexity, Google's AI Mode, agentic browsers, plus a long tail of voice agents, WhatsApp bots, vertical AI search tools and funded startups going straight for the dealer's wallet. Cumulatively, it adds up fast.

Part three — the precedent

This is the part I find genuinely hopeful, and it comes from outside the car industry.

Walt Disney drew a diagram in 1957 — TV, film, music, parks, merchandise, all separate businesses, all feeding one another. Distributed channels, centralised value. Every studio that didn't think that way is gone. Disney is worth around $200bn today.

Lego lived the same lesson the hard way. In 2003 they were $800m in debt, losing a million dollars a day, eighteen months from bankruptcy — undercut by imitators and out-competed for kids' attention by video games. Sound familiar? They didn't fight the disruptors. They absorbed them. By 2024 they were the biggest toy company in the world, $10.8bn in revenue.

The same shape applies to marketplaces: open the data, route the leads, share the revenue, keep the centre.

The strategic choice

That leaves three honest postures — none of them obviously wrong:

Build your own. Own the UI and the data end-to-end. Defensible, but you build alone and the consumer still has to come to you.
Partner with a platform. Distribution at the speed of ChatGPT — but the user relationship sits with the platform, not with you.
Open the interface. Become an agent-readable layer that every agent queries — the default plumbing for the category. You trade exclusivity for ubiquity.

There's an asymmetry worth naming: opening up is genuinely hard for the #1 player, because every open interface risks cannibalising first-party traffic and decades of dealer relationships. The defensive bet is the rational one for the incumbent. But for the #2, the maths flips — there's less first-party share to protect, distribution beats reach when your reach is small, and becoming the default plumbing locks in the next generation. As one operator put it: "LLMs power the intelligence, but the edge comes from domain-specific data."

So I built one to prove the point

The live build — askautosbuzz — is a natural-language car-search agent sitting on top of real automotive data. It's the kind of "product layer" that used to take a funded team. It now takes an afternoon. That's the whole argument in one artefact: the intelligence is commodity; the data and the relationships are the moat.

The question I left the room with is the one I'll leave you with: the future of this gets built either way. The only question is whether you build it.

Where to find it

Everything lives in one place:

🎬 Watch the full talk (~15 min): youtube.com/watch?v=uNqBh1PHEJ4
📑 Read the slides and 🤖 see the live build: askautosbuzz.com
✉️ Want to talk it through? There's a contact form on the page, or reach me directly at alastair@hurricane.works.

If you're building in this space — or trying to work out which of those three postures is right for your business — I'd genuinely like to hear from you.