Trajectory launches continuous-feedback startup

Why it matters

Trajectory emerged on May 27 with a pitch that cuts against how most AI products are built today: models should keep improving after launch instead of freezing once training ends. Wired reported the startup was founded by former Google and Apple researchers and is building systems that learn from live user feedback in production. The company’s thesis is that deployment data — not just pretraining data — should shape how models are updated over time. ### Why are the founders focused on feedback after deployment? Wired’s May 27 report said Trajectory is trying to build what it called AI’s “missing feedback loop,” a way for products to learn from real user interactions once they are already in use. That addresses a common pattern in current AI deployment, where systems are trained, shipped and then only updated in larger, slower model releases. (boulderdaily.net) Ronak Malde, identified in follow-on reports as Trajectory’s chief executive and co-founder, said the company wants to apply the fast iteration seen in AI coding products to a broader set of enterprise tools. Other reports citing the launch said the founding team also includes researchers with experience at Apple, Google DeepMind, OpenAI and Meta. (boulderdaily.net) ### What problem is Trajectory saying current AI products still have? Trajectory’s argument is that many deployed models do not systematically learn from their own failures in production. In practice, that means a customer-support bot, internal assistant or workflow agent may generate logs, corrections and user reactions every day, but those signals are not always turned into rapid model improvements. (digitrendz.blog) Wired framed that gap as the core opening for the startup. Secondary reports summarizing the story said Trajectory plans to use company-specific interaction data to identify failures and feed those results into recurring updates, rather than waiting for occasional retraining cycles. ### How is this different from the usual “better model” startup pitch? (ainovat.com) Trajectory is not being described primarily as a frontier-model lab. The reporting around the launch points instead to infrastructure and workflow: collecting feedback, evaluating failures, and pushing improvements back into deployed systems. That places the company closer to the layer that sits between users and models than to companies competing to build the largest base model. (boulderdaily.net) That framing also matches a broader shift in the AI market toward orchestration, evaluation and routing. The company is betting that the operational loop around a model — how it is measured, corrected and updated — can matter as much as the underlying model itself. That is an inference from the launch framing and the product description in the reporting. ### Why does “continuous evaluation” matter in this pitch? (ainovat.com) Continuous evaluation is the mechanism that makes a feedback loop usable. A company first needs to detect where an AI system failed, whether users corrected it, and whether a later update actually improved the result. Without that layer, feedback remains anecdotal rather than operational. Reports following the launch said Trajectory plans to help companies ship updates on a recurring basis, including weekly model improvements in some use cases. (ainovat.com) That suggests the startup is selling a process as much as a model: gather signals, score behavior, retrain or post-train, redeploy, and measure again. ### What should readers watch next? (ainovat.com) Wired’s May 27 story provides the clearest early marker: whether Trajectory can turn live user interactions into repeatable updates for real products outside coding tools. The next concrete signs will be named customers, disclosed product workflows, and any details the company releases on how often it can push improvements from production feedback. (digitrendz.blog) Other reports tied to the launch said Trajectory had raised seed funding and attracted investors including Jeff Dean and Fei-Fei Li, but those details were not available from the primary Wired excerpt surfaced in search, so they should be treated as follow-on reporting rather than the core confirmed launch facts here. (digitrendz.blog) (boulderdaily.net)

Key numbers

• Trajectory launched on May 27, 2026, with former Google and Apple researchers pitching software that updates AI products from live user feedback.
• Wired’s May 27 report names Ronak Malde among the founders and says the company is targeting continuous evaluation workflows.
• Trajectory emerged on May 27 with a pitch that cuts against how most AI products are built today: models should keep improving after launch instead of freezing once training ends.
• Wired’s May 27 report said Trajectory is trying to build what it called AI’s “missing feedback loop,” a way for products to learn from real user interactions once they are already in use.

What happens next

• Trajectory emerged on May 27 with a pitch that cuts against how most AI products are built today: models should keep improving after launch instead of freezing once training ends.
• Wired’s May 27 report said Trajectory is trying to build what it called AI’s “missing feedback loop,” a way for products to learn from real user interactions once they are already in use.
• Other reports citing the launch said the founding team also includes researchers with experience at Apple, Google DeepMind, OpenAI and Meta.

Sources

• boulderdaily.net
• digitrendz.blog
• ainovat.com

Quick answers