Commoditized Magic & Agentic Marketplaces: What’s Next for Founders and Investors | Yariv Adan

What we talked about.

In this episode of the Sunny Ray show, host Sunny Ray talks with Yariv Adan of Ellipsis Ventures about two ideas he believes will reshape startups and venture investing: commoditized magic and agentic marketplaces. Adan, who spent close to 20 years at Google working on Google Assistant, Google Lens and Google Cloud AI, explains how AI has made once-rare capabilities both transformative and available for about $20 a month. He argues the core rule of VC still holds, since businesses need defensible differentiation, but the bar for it has risen sharply. Adan predicts that once agents begin buying software, procurement will shift from trust to proof, with agents bidding against each other in open marketplaces and squeezing margins. He names data, expertise and non-trivial compute as the remaining sources of lock-in, and cautions founders against merely automating the past. The conversation closes with a look at the first signs of this shift in tools like Codex and Claude Code, and an overview of Ellipsis, an early-stage AI deep tech fund.

When AI magic costs $20 a month and agents do the buying, what can founders still build that nobody can copy?

The questions, and the answers.

What do commoditized magic and agentic marketplaces mean in practice, and what should founders and investors watch?

Commoditized magic means AI delivers capabilities that used to be nearly impossible, yet it is available to everyone for about $20 a month. The VC rule of defensible differentiation still holds, but the bar is much higher. Agentic marketplaces are the coming earthquake, where agents buy software instead of humans, and I think that arrives within about two years.

If AI makes the magic available to everyone, what can a startup still build that competitors can't copy?

First, don't automate the past, unlock the future. In SaaS automation I struggle to bet on winners. I look for new lock-ins: proprietary data, expertise others cannot easily have, and maybe non-trivial compute that is not commoditized. Some bets are speculative exits, since big companies need acquisitions. But finding true defensibility is easier said than done.

How would an agentic marketplace actually work?

You describe your goal, and the system asks about price, timing, location, security and privacy. Then it bids out the task to agents that can solve it. They run inside your sandbox, so you do not move your data or install anything. Stickiness is close to zero, so undifferentiated players end up fighting on price in every transaction.

How does buying change when agents are the buyers?

In the human world you rely on trust, because you cannot meaningfully verify a vendor. Agents can replace trust with proof. An agent can reach a million other agents, ask them to demonstrate they can do the job, run your eval examples, or point to verifying customers. I also expect experts to guarantee agents and take part of the transaction.

What is the first real sign that agents buying software is happening?

It is already happening in Codex and Claude Code, through connectors and MCP. I think the model providers want to be paid on the value of the transaction rather than tokens, since token pricing is a bad model for them. Users are also moving from approving everything to auto mode, and players like OpenRouter are starting to build marketplace pieces.

Why is this a harder business for VCs?

A marketplace like this commoditizes the economy, so it becomes much larger, but margins get squeezed, and it is unclear that economies of scale help. Many sectors, like law offices and accounting, never attracted VCs. Many investors have billions in the old world, so realizing this is changing is a hard thing to accept.

What is Ellipsis and which founders should reach out?

Ellipsis is an early stage AI fund I built with a partner who has an ETH PhD and led engineering at Apple vision intelligence. We invest in AI deep tech companies solving very large problems with clear data and differentiability. We love being first check and are very active, supportive investors, with strong networks in big tech, corporate and VC.