Startup raises $45M to let nonexperts teach AI their jobs

Trajectory, a startup about a year old, has raised $45 million to build a tool that lets specialists teach AI models to do their jobs better. The round values the company at $300 million. Sequoia led it, and Nvidia, Bessemer Venture Partners and Bain Capital Ventures took part. The company has 18 employees.

The problem it targets

Today's AI models, whether from OpenAI, Anthropic or open-source projects, do not learn from their mistakes unless they get more training. Businesses in fields like law and finance usually deal with this by fine-tuning, which means adjusting a general model for a specialty. That often takes an in-house research team or expensive consultants, known as forward-deployed engineers.

How Trajectory works

Trajectory wants to automate that work through an approach called continual learning. A business starts with an open-source AI model and connects Trajectory's platform to its apps. The platform collects feedback on how the model performs, then uses it to retrain the model so it gets better at specialized tasks over time.

Co-founder Ronak Malde compared today's models to a very smart PhD student who is always starting a new job. "But what the economy values is someone who worked on the job for 30 years," he told TechCrunch. He said lawyers, doctors and other specialists can improve the models near automatically themselves.

Who is using it

Harvey, a legal AI firm, and Rogo, a finance AI firm, already use the platform to train open-source models. The goal is to make them act more like experienced lawyers or financial analysts. Examples include reviewing a certain type of legal contract or valuing a company with a specific method. Other customers include Decagon, Mercor and Nvidia. The platform is now publicly available.

Safety and competition

Trajectory's method resembles recursive self-improvement, the idea that AI systems improve their own abilities. Co-founder Michael Elabd argues that keeping humans in the training loop is safer, because fully autonomous self-improvement could lead to unintended or even dangerous results.

Rivals exist. Trajectory names Thinking Machines Lab, which makes the fine-tuning platform Tinker, as a direct competitor. It also points to Applied Compute, valued at $3.25 billion last month, and Prime Intellect, which helps companies build and train their own AI agents.

What is not known yet

The material gives no pricing and no results showing how much the models improve. It also does not say how well non-experts manage without technical help. Trajectory says its interface is designed to make training easier for nontechnical customers, but that claim has not been tested in public here.