A $1 billion Series A: what Mirendil's round says about where AI money is going

Mirendil raised $1 billion for AI research agents in a week when most big rounds went to agents or the infrastructure under them. What product teams should take from it.

FluxOrbit3 min read

A Series A used to be the round where a startup proved it had a business. Last week a San Francisco company called Mirendil raised one worth $1 billion.

The round was co-led by Kleiner Perkins and Clearlake Capital, with Google, Battery Ventures and several others taking part. Counting an earlier $200 million seed, Mirendil has now raised about $1.2 billion. It has not disclosed a valuation. Its main product is still in private beta.

It is easy to roll your eyes at a number like that. We think it is more useful to look at what the money is being asked to do, and at what else got funded in the same seven days.

What Mirendil is building

Mirendil makes tools that let a single scientist or a small engineering team run research with several AI agents at once. Its product, Lor, coordinates agents that write code, run experiments and keep working across long research sessions. The founders are researchers who previously worked at large AI labs.

The company's headline claim is that one researcher working with its agents trained a coding model, Mirendil-Base, in three and a half months on less than $10 million of compute. That figure comes from Mirendil and has not been independently verified, so treat it as a pitch, not a result.

The pitch is clear enough, though. If agents can do part of the experimental work, a research team's output stops being tied to its headcount. Investors are betting a billion dollars that this is true, and that whoever builds the coordination layer gets paid for it.

The rest of the week pointed the same way

Crunchbase News tracks the ten largest US rounds each week. For 7 to 13 November it says the vast majority were AI-related. A few of them:

  • Mercor, $750 million at a $20 billion valuation. It connects AI developers with human experts who train and evaluate models.
  • Torus, $500 million in growth financing led by Magnetar Capital, for distributed power and compute infrastructure.
  • LangChain, $300 million at a $3 billion valuation, led by CapitalG. Its tools test, trace, monitor and deploy AI agents. The company was valued at $1.25 billion in October 2025.
  • Path Robotics, $300 million at $2.45 billion. It builds AI for manufacturing robots, starting with welding.
  • Nominal, $240 million at a $3 billion valuation, for software used by hardware engineering teams.

Outside the US, Sydney-based Q-CTRL raised US$290 million, led by Salesforce Ventures, for software that controls quantum computers and sensors. Further down the table, Kindo raised $30 million for a product that lets companies see which AI agents they are running, what data those agents can reach and which rules apply to them.

Put those side by side and three themes show up.

Agents are getting their own tooling. LangChain, Kindo and Mirendil are all selling some version of "run many agents and keep them under control". Not long ago, many teams built that layer themselves.

The bottleneck is physical. Torus is funded to find power for compute. Money is going into the things that limit AI, not only into the models.

The big numbers are very concentrated. TechStartups added up the rounds announced on 12 and 13 November and found that three companies, Mirendil, LangChain and Q-CTRL, accounted for about 84% of the total. A few enormous cheques do not mean money is easy for everyone.

What we would take from it as builders

If you are putting AI agents into a product this year, two practical points.

First, do not build your own tracing, evaluation and permission system unless that is your product. Funded companies are now competing to do it for you, and their tools will improve faster than an internal side project.

Second, ask the governance question before you ship, not after. Products like Kindo exist because it is easy to end up with agents that can reach data nobody signed off on. Know which agents you run, what each one can touch and who approved it. That costs little at the start and a great deal later.

As for the billion-dollar Series A: it tells you where investors think the next few years are going. It does not tell you the product works. We will be more interested when Lor leaves private beta and people outside the company can try it.

Building something that has to stay up and stay safe?

We design, build and look after software for teams that cannot afford surprises.

WhatsApp