Don't Build AI Agents. Build a Network of Them.

01 /The Concept

This startup was founded in Spain earlier this year. It hasn't even publicly launched its platform yet – but it's already raised $10M in its first funding round. Investors clearly found the idea compelling enough.

Causa Prima is building a network of AI agents for the finance and procurement functions of companies that buy and sell digital products, goods, and services.

As the startup puts it, we keep hearing stories about how AI will simplify business. But in its simplest form, that promise is a trap.

Here's why: when the cost of something drops, people use more of it. This observation was formalized back in 1865 as Jevons' Paradox. It's been confirmed repeatedly – cheaper mobile data led to ubiquitous smartphones, cheaper ride-hailing led to everyone taking Ubers, and so on.

What happens when AI agents enter procurement and supply chain workflows? Imagine an AI agent at a buying company that floods every possible supplier with requests to participate in every auction. Suppliers would drown in requests they can't process. The same dynamic applies in reverse: suppliers blasting every possible buyer with offers would drown procurement teams.

The efficiency gains from faster, cheaper, higher-volume AI operations would produce such an explosion of inter-company communications and transactions that humans simply couldn't keep up.

This makes it largely pointless to deploy AI agents on just one side of the transaction. If your agent is fast but the counterparty is still human, the interaction runs at human speed – and you've gained nothing meaningful.

The right model isn't isolated AI agents. It's a network of AI agents deployed simultaneously across all parties. Then agent talks to agent at machine speed and machine volume.

And those interactions don't need to be simple. Agents can negotiate, agree on terms, set payment schedules, request early-payment discounts, and execute transactions – with humans stepping in only for final decisions or exceptions.

Causa Prima is building exactly this network, focused on the financial layer: discussing, agreeing, and settling payments between companies. The result: faster, lighter, more efficient finance operations on both the buyer and supplier side.

The platform isn't publicly live yet, but it's already running in pilot with 3,000 finance teams across various companies.

02 /Why It Matters

A month ago, a [related review](/review/pora-vzryvat-zakupki) covered ConstructionBevy – a platform for streamlining auctions between developers and contractors. That startup built tools that worked independently on each side of the transaction.

The review noted then: "This problem needs to be solved on both sides simultaneously – contractors/suppliers and buyers. Otherwise it will eventually break down."

And: "The ideal picture is an 'AI auction' where the buyer's agents send out proposals and the contractor's agents sort through them, communicating with each other to clarify key points. Humans only step in once the agents have agreed on everything. ConstructionBevy is only a first approximation of that ideal."

Causa Prima is getting much closer to that ideal – though for now, only on the financial transactions side.

But the potential scope is wider than procurement. In March, a [related review](/review/k-tvoej-polze-i-vygode) covered Tobira – a professional network where AI agents represent their human owners. A founder's agent can meet an investor's agent, pitch the startup, and receive a pitch in return. If both agents judge the match promising, they notify their owners. Only after both humans confirm interest do the agents exchange contact details – until then, both parties remain anonymous to each other.

The result is dramatically faster relationship formation, since agents can establish contacts at speeds no human could match – and without the awkwardness of cold outreach.

For some startups, building the network around a two-sided model seems almost unavoidable. Financial News Systems ([related review](/review/dva-poka-eshhjo-nelegalnyh-rynka)), which raised €1.5M in March, built an AI platform that publishes investment-critical financial news with an average delay of 13 milliseconds from event to publication.

But as that review observed: "If the news arrives in milliseconds, the decision based on it also needs to happen in milliseconds. Otherwise what's the point of the speed? And who can decide in milliseconds? Not humans – only AI agents." Which means the natural next step for Financial News Systems is to let subscribers create AI agents directly on the platform, capable of acting on published news at the same inhuman speed.

03 /Opportunities

The direction: build networks of AI agents where agents interact with each other, simultaneously serving all parties in a process.

The key capability requirement: agents that can negotiate, evaluate, discuss, and agree – not just execute mechanical tasks – so that humans are removed from the loop entirely or limited to final sign-off.

The appeal of this model, once it works, is that it scales extremely fast. Either the network effect kicks in and pulls in the rest of the market participants... or it doesn't.

So – which market niche would you want to cover with a network like this?