Startup Lets Firms Refine AI on Live Usage Data
Trajectory's self-serve platform, live since November 2, 2026, turns user corrections into training signal for deployed models.
Updated on Nov. 2, 2026 in Artificial Intelligence
Live Poll
Would you be comfortable with companies training AI on your edits and corrections?
Trajectory, a startup founded by former researchers from Google DeepMind, OpenAI and Apple, launched a self-serve platform for post-training AI models on production data on November 2, 2026. The development is shipped: the platform is live, offering a lightweight software development kit that captures user edits and corrections to AI-generated output.
Why it matters
The platform turns everyday product usage into training signal, so companies can refine deployed models on their own production data rather than relying only on static datasets. That points toward a model-improvement loop that runs continuously in production.
The figures are company-level financing and valuation numbers, with the valuation rising from $115 million at seed to $300 million in the Sequoia-led round. The extraction includes no benchmark or measured model-performance result for the platform.
The players
Trajectory
An AI post-training startup building a software development kit that captures user corrections to feed model refinement, with customers including Clay, Harvey, Decagon, Mercor and Rogo.
Sequoia
A venture capital firm that led a round valuing Trajectory at $300 million.
The details
Companies plug the Trajectory software development kit into their products to record usage signals such as edits and corrections a user makes to AI-generated output. Those signals feed post-training cycles, the step where a model is further adjusted after its initial training, and these cycles can run weekly or more often. A feature called continual grading is designed to scale human preferences, meaning it aims to turn people's judgments of outputs into a repeatable evaluation signal.
Timeline
Trajectory exited stealth mode in May 2026.
The self-serve platform went live on November 2, 2026.
The Tech Race
Trajectory positions its SDK as a way for companies to capture corrections and feed them into post-training, a niche distinct from building base models. The extraction does not name a specific competing product or research program to compare it against.
Companies that integrate the Trajectory SDK can have end-user edits and corrections captured and fed back into model updates, potentially changing how quickly their AI features improve. Pricing, integration timelines and the scope of available features beyond the SDK are not stated in the source.
The takeaway
The platform's weekly or faster post-training cycles suggest deployed AI products may increasingly improve from their own usage data. Watch for published performance results from customers using continual grading to confirm whether captured corrections measurably improve model quality.
Further reading
For background on how these systems are refined after deployment, see Artificial Intelligence.
Source note: This article includes information reported by Crypto Briefing.
Live Poll
Would you be comfortable with companies training AI on your edits and corrections?
