Trajectory, an 18-person startup founded about a year ago by three Stanford classmates, has raised $45 million in a Series A round at a $300 million valuation to automate the post-training of AI models. The round was led by Sequoia, with Nvidia, Bessemer Venture Partners, and Bain Capital Ventures joining, and the company is now making its platform publicly available.
The problem Trajectory targets
General AI models share a critical limitation: they do not learn from their mistakes without further training. Businesses that adapt those models for law, finance, or medicine usually fine-tune them, but that work typically requires hiring internal research teams or relying on expensive consultants known as forward-deployed engineers.
Trajectory co-founder Ronak Malde compared a foundation model to a very smart doctoral student who is always starting a new job. What the economy values, he told TechCrunch, is someone who has worked on the job for 30 years. The startup wants to close that gap by letting models accumulate experience from real work, rather than restarting with every task.
Its approach is continual learning. A business starts with an open-source model, connects Trajectory to its applications, and collects feedback on how the model performs. Trajectory then uses that feedback to retrain the model, helping it improve at specialized tasks over time, without the customer building a research team.
Who is already using it
Harvey and Rogo already use the platform to post-train open-source models, helping them perform more like experienced lawyers or financial analysts. They learn from previous tasks, such as reviewing a particular type of legal contract or valuing a company using a specific methodology, rather than from generic examples alone.
Trajectory’s other customers include Decagon, Mercor, and Nvidia, and the people using the system decide what good looks like for their organization. That judgment is then turned into improvements for their model, Malde said, so the definition of quality stays with the specialists, while the retraining machinery sits with Trajectory.
The founders bring relevant backgrounds to that pitch. Malde was among the researchers Google DeepMind acquired from Windsurf, co-founder Michael Elabd also worked at DeepMind, and third co-founder Arjun Karanam previously worked on generative foundation models at Apple. The team has also recruited researchers and product designers from OpenAI, Stripe, and Netflix.
Why It Matters
If post-training becomes a service that specialists can run themselves, the cost of deploying AI in regulated, expert work could fall sharply. A law firm or clinic would no longer need to choose between a generic model that misses local practice and a bespoke build that only a research team can maintain. That middle path is where Trajectory is aiming.
The competitive field is crowded, and Trajectory names it openly. It considers Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati and maker of the Tinker fine-tuning platform, a direct competitor, along with Applied Compute, valued at $3.25 billion last month, and Prime Intellect, which helps companies build and train their own AI agents.
Our read
Trajectory’s pitch is strongest where it is most modest: it does not claim to invent a new model, but to remove the engineering bottleneck between a general model and a useful specialist. That bottleneck is real, and Harvey and Rogo give the claim weight, because legal and financial work punishes vague answers a demo cannot hide.
The open question is how far continual learning can go without drifting. Feedback loops can entrench a firm’s habits, including its mistakes, unless graders stay rigorous and retraining is checked against held-out cases. Trajectory’s human-in-the-loop answer is reassuring, but the company has not published the guardrails that would let outsiders judge that risk.
Sequoia leading a $45 million round at a $300 million valuation, with Nvidia alongside, signals that investors see post-training as infrastructure rather than a feature. If that bet is right, the winners will be platforms that earn trust from specialists first, not those that promise autonomy fastest. Trajectory’s public launch is a test of that thesis.
Background
Trajectory’s approach shares some similarities with recursive self-improvement, the idea that AI systems can improve their own capabilities. Co-founder Michael Elabd argues that keeping humans involved makes Trajectory’s approach safer, since fully autonomous improvement could produce unintended or even dangerous outcomes.
Source: TechCrunch report on Trajectory
0 Comments