Making Agents Work
Announcing Fern Labs: Agent performance at the scale of real-world problems
We’re excited to launch Fern Labs, an AI-first company focused on deploying large coordinated agent systems to solve real-world problems. In this piece, we’ll share why this work matters, the challenges we’re tackling, and our immediate roadmap.
The Agent Landscape in 2025
Over the past year, large language models (LLMs) and their multi-modal derivatives have rapidly increased in capability while inference costs are dropping by a factor of ten or more every year. Together, this is unlocking more ambitious and complex AI use cases than ever before.
The challenge is to provide these agents with flexible environments where they can collaborate at scale while remaining grounded in real-world needs. Businesses must keep apace with these evolving capabilities without having to reinvent the wheel every time a new model, agent, or product emerges.
Why Fern Labs?
We founded Fern Labs because we believe massive, autonomous, coordinated information systems are coming this year, and they must be built in the field. At Palantir, we built early agent-driven systems and saw firsthand how they could reshape how companies operate. But achieving scalable performance across the vast networks of agents that every enterprise will need demands a huge amount of R&D. Solutions developed in a vacuum often look promising but tend to fail in real-world production.
Our goal is to build Fern - a maximally powerful system for orchestrating and deploying the first (of many) generations of truly useful agents. By applying the principles of forward-deployed engineering to large-scale agent R&D, we ensure we’re building systems that truly work for - and with - people.
The technical infrastructure underlying Fern is designed to be:
- Future-proof – We build with the assumption of continued model improvement
- Robust – Results must be reliable, performance must be scalable
- Flexible – Agents, with configuration, should be able to handle any complex digital task
- Continuously integrated – We track the latest agent offerings to incorporate best-in-class technology and tooling
The Fern platform will accelerate agent-driven problem solving - from small proof-of-concepts to large-scale solutions that address problems currently requiring entire companies.
A New Type of Company
Today we’re at a stage where an agent can create a simple app or perform a straightforward data science project. This year, we expect to see agents that can independently generate medium-sized software projects - completely on their own. We will see a shift from demos to genuinely helpful, production-ready tools.
As this evolution accelerates, we anticipate a radical change in how companies are structured. The traditional approach has a team painstakingly building every feature by hand. We’ll instead see a new model emerge whereby a company builds a machine (soon to be Fern) that builds its products.
This is similar to how trading floors shifted from manual decision-making to high-frequency algorithms that devise and execute trades in microseconds. The same principle applies: what today takes months or years could soon happen in days, hours, or minutes, executed by thousands of coordinated agents.
Fern Labs is working to construct that “machine”. Our R&D centres on running networks of agents that can tackle wide-ranging tasks, respond to new information quickly, and scale up to meet any challenge. We are application agnostic: our systems can be used to help create software, find drug candidates, model complex financial scenarios, or crunch terabytes of data for insights in data science.
Moving Forward
With Fern Labs, we aim to help usher in a new era of massive autonomous coordinated information networks. This will lead to huge empowerment for the individual, as solving a problem becomes akin to having a conversation.
We’ve started by tackling smaller projects, and are gradually working up to more significant challenges through close collaboration with our design partners in cancer research, finance, and data science. Each partnership helps us refine Fern, ensuring it can handle real-world complexity and deliver tangible value.
In the longer run, our work will empower hyper-reactive companies and research institutions that solve problems more quickly - and more creatively - than ever before.
We’re excited to drive this journey and look forward to sharing more in the weeks and months ahead. If you’d like to learn more or explore potential partnerships, feel free to reach out at contact@fern-labs.ai.
Welcome to Fern Labs - making agents work.