AI agents that actually shipFintech with real marginsCustom software, not off-the-shelfResearch that reaches productionBacking people first, slides secondReal estate with a tech lensFairways over pitch decksSkin in the game — alwaysSignal over noiseBerlin-rooted. Worldwide reach.
André Beyer

Investing

Agentic Systems Investing

I invest through Beyer Ventures, at pre-seed and seed. Vertical AI and agentic systems are two of the three areas I understand well enough to write early checks into. Here is what I mean by those terms and what I look for.

What I mean by agentic systems

An agentic system is software that takes a goal, decides on the steps, uses tools to carry them out, and checks its own results. The difference from a chatbot is that it does the work instead of describing it. The difference from classic automation is that the steps aren't fixed in advance.

That definition is broad, and most products that carry the label today are a model call with a loop around it. That's fine as a starting point. What I'm interested in is the part that comes after: the system runs unattended on real work, someone is accountable for the result, and the company knows how often it's right.

The term is also overused in pitch decks. I don't invest in the label. I invest when the agent does a job that someone currently pays for and the team can show it doing that job.

Why vertical

Vertical AI means building for one industry or one function, deeply. Not an assistant for everyone, but for example a system that handles claims for insurers, plans appointments for clinics, or answers tenders for construction firms.

I prefer vertical for three reasons. The workflows are specific enough that an agent can be evaluated — you can say what a correct outcome is. The data, integrations, and domain rules take real effort to get right, and that effort is hard to copy. And the buyer already has a budget for the work, because people are doing it today.

Horizontal tools compete with the model providers and with every other team that has access to the same APIs. A vertical team competes on how well it knows the customer's work. I'd rather back the second kind.

The third area on my list is automation infrastructure: the tooling that agent systems need to run reliably. Same logic — a specific problem, owned by people who've had it themselves.

What Beyer Ventures invests in

The facts, as they stand on the Beyer Ventures startups page:

  • Stage: pre-seed and seed.
  • First checks from EUR 25,000 to EUR 200,000, with follow-on capacity for companies we back.
  • Focus: vertical AI, agentic systems, automation infrastructure — and founders with a problem they clearly own.
  • Where: Berlin and internationally. If the fit is mutual, geography is not the blocker.
  • A small number of deliberate investments rather than a wide portfolio.

Conviction and depth matter more than the sector label. If what you're building doesn't fit neatly into one of the three areas but you clearly own the problem, that counts for more than the category.

The teams I back

I invest in people first, slides second. Three things get my attention:

  • A warm introduction from someone I trust. Not as a gate, but because it tells me something a cold email can't.
  • Signal in the work: shipped product, real traction, or genuine technical depth. One of them is enough at this stage. None of them is not.
  • A shared obsession. You care deeply about something I also care about, so the conversations are worth having for both of us.

Beyond that I look for founders with conviction, real technical depth, and a problem they've clearly owned — ideally because they had it themselves.

What I ask about an agentic product

When I look at a company in this space, the same questions come up every time:

  • What job does the agent do, and who did it before? If the answer is nobody, I want to understand why the work matters.
  • How do you know it works? I'm looking for a test set from real usage and a failure rate you can state, not a demo.
  • What happens when it's wrong? Who notices, how fast, and what does the mistake cost the customer?
  • What does a run cost, and how does that compare to what the customer pays? Margins in agent products can disappear in model bills.
  • What do you have that the next team with the same model doesn't? Usually the answer is data, integrations, or distribution in one industry.
  • What happens to the product when the underlying model gets twice as good? The best answer is that your product gets better too.

You don't need perfect answers at pre-seed. I do want to see that you've asked yourself the same questions. The second one is a topic of its own; I've written about it under AI agent evaluation.

What you can expect

A fast decision. No polite maybes and no empty promises — you get a yes or a no. A first check within weeks, not quarters.

An operator's perspective, not only a check. I've built and led tech companies in Berlin for a decade, as CTO, as CEO, and as an investor. That's what I bring when the question is about architecture, the first technical hires, or how to get from pilot to contract.

Follow-on capacity for the companies I back, and a long holding period. What I expect in return is simple: you own your problem, you make progress, and you keep me in the loop. Bad news early beats good news late.

What it isn't

It isn't agency work with equity attached. If what you need is someone to build your product for you, an investment from me is the wrong tool.

It also isn't spray-and-pray. I make few investments and spend time on each. That means I pass on good companies regularly, including ones where the only problem is that I don't understand the domain well enough to be useful.

How to start

You don't need a pitch deck. A few honest lines about what you're building and how we're connected are enough. The form is on beyerventures.com/startups.

Investing is one of four ways I work with founders; the others are on Work with me. If you're unsure which applies, the FAQ answers the common questions. And the homepage gives you the short version of who I am.

Related