Trajectory Raises $15M to Build Continual Learning AI Infrastructure
Trajectory raised $15M to build continual learning infrastructure for enterprise AI systems that improve from real-world user interactions.
Artificial intelligence has a memory problem, and the industry has been pretending otherwise for far too long. A model aces the benchmark. The demo goes viral. Venture capital firms start talking about “agentic AI” workflows like they just discovered fire behind a Sweetgreen. Then the product hits real users. Customers correct outputs, retry prompts, edit responses, and reject recommendations. The AI absorbs a mountain of behavioral signal, yet most systems never actually learn from it after deployment. That gap between intelligence and adaptation is where Trajectory just planted a very expensive flag.
Trajectory, a San Francisco-based AI infrastructure company focused on continual learning and post-deployment AI systems, announced $15M in Seed funding led by Conviction, with participation from Bessemer Venture Partners, Radical Ventures, BoxGroup, and angel investors Jeff Dean and Fei-Fei Li. The round reportedly values Trajectory at $115M post-money and positions the company inside one of the fastest-emerging categories in enterprise AI infrastructure: systems that continuously improve from production usage instead of remaining static after deployment.
The company was founded by Ronak Malde, CEO and Co-Founder, alongside Co-Founders Michael Elabd and Arjun Karanam. Their backgrounds trace through Google DeepMind, Apple, robotics research, Vision Pro, and Windsurf’s AI infrastructure work. Different technical disciplines. Same conclusion: production AI still behaves like software with short-term memory loss.
What Happened
Trajectory publicly launched with a platform designed to help enterprise AI systems continuously improve after deployment. The company’s infrastructure captures real-world usage signals including edits, retries, corrections, traces, and re-prompts through a lightweight SDK integrated directly into products. That sounds deceptively simple until you realize how much enterprise AI still operates like an expensive science project wrapped inside executive optimism.
A surprising amount of enterprise AI infrastructure still relies on static training cycles while users generate massive amounts of behavioral feedback that disappears entirely. Models are trained once, deployed broadly, monitored nervously, then patched manually when things drift sideways in production. Trajectory wants to operationalize that feedback loop.
The company structures its workflow around 4 stages: Instrument, Understand, Steer, and Learn. The broader thesis is clear: deployed AI systems should evolve continuously from user interactions rather than remain frozen snapshots trained exclusively on static datasets. That distinction matters more than the market currently admits.
Why Continual Learning Matters
The AI market spent the last several years obsessed with scale. Bigger models. Larger context windows. Faster inference. More GPUs stacked like poker chips inside hyperscale data centers. Now enterprise operators are running into a different reality: static intelligence degrades quickly in production environments.
Benchmarks look polished in research papers. Customers do not behave like benchmarks. Real users phrase requests poorly, change objectives halfway through workflows, and introduce edge cases no evaluation suite anticipated. They correct systems thousands of times per day without realizing they are generating some of the most valuable training data inside modern software infrastructure.
Trajectory is betting that continual learning infrastructure becomes foundational for the next generation of enterprise AI systems and agentic AI applications. That thesis aligns with a broader shift happening quietly underneath the louder AI-agent conversation. Sophisticated operators increasingly recognize that autonomous systems without adaptive learning eventually become brittle systems. Intelligence without iteration has a shelf life. The market is beginning to separate AI demos from AI infrastructure.
The Team Behind Trajectory
Trajectory’s founding team reflects a broader migration pattern happening across the AI ecosystem, particularly among researchers focused on adaptive systems and post-training infrastructure. Ronak Malde previously worked in AI research at Windsurf before moving to Google DeepMind following the acquisition. Michael Elabd worked in robotics at Google DeepMind. Arjun Karanam worked on Vision Pro-related AI systems at Apple.
That combination matters strategically. Robotics researchers think differently about learning loops because physical systems fail publicly and immediately. Vision Pro development forces teams to handle multimodal interaction complexity in real-time environments. DeepMind’s research culture pushes aggressively toward adaptive systems capable of iterative improvement.
Put those disciplines in one room long enough and eventually somebody starts building infrastructure for AI systems that actually learn after deployment instead of pretending benchmarks tell the whole story. The investor syndicate also signals where sophisticated capital believes the AI stack is heading. Conviction, Bessemer Venture Partners, Radical Ventures, and BoxGroup have consistently concentrated around infrastructure-layer AI companies rather than thin application wrappers chasing temporary hype cycles. Jeff Dean and Fei-Fei Li participating adds additional technical gravity to the round. Serious technical investors increasingly appear more interested in infrastructure durability than surface-level AI theatrics.
Governance Is Becoming a Competitive Advantage
One of the more important signals inside Trajectory’s launch is the company’s emphasis on governance and deployment oversight. Trajectory states the platform is SOC 2 Certified and includes approval workflows before deployment, auditable training runs, and customer-controlled training boundaries. That may sound procedural. It is not.
Enterprise AI adoption is colliding directly with operational trust issues. Companies want adaptive systems, but they do not want autonomous chaos ricocheting through customer workflows at 2:13 a.m. because a model decided to become “creative” during retraining.
The enterprise AI infrastructure market is entering a phase where governance becomes part of product differentiation rather than simple compliance overhead. That shift mirrors earlier cloud infrastructure cycles where observability, security controls, and deployment reliability eventually became mandatory buying criteria instead of optional product features. Trajectory appears to understand that dynamic early.
What This Signals About the AI Infrastructure Market
Trajectory’s reported $115M post-money valuation only days after public launch says something important about where sophisticated capital believes the AI infrastructure market is heading. The infrastructure layer is still forming.
The market conversation remains heavily focused on applications, copilots, and AI agents because those products demo well and generate immediate attention. Infrastructure companies operate differently. Their importance compounds slowly until entire ecosystems depend on them.
Datadog looked boring before observability became mission-critical. Snowflake looked niche before data infrastructure became strategic leverage. NVIDIA looked like a gaming company before AI training consumed modern compute markets. Trajectory is positioning itself around a similar long-cycle thesis: AI systems that continuously learn from production environments may eventually outperform systems trained primarily through static pre-deployment methods.
That is not just a product argument. It is a philosophical argument about how software evolves. Right now, the companies building adaptive AI infrastructure increasingly look more important than the companies merely packaging intelligence into prettier interfaces.
Frequently Asked Questions
What is Trajectory?
Trajectory is a San Francisco-based AI infrastructure company building continual learning systems that help deployed AI models improve from real-world user interactions.
How much funding did Trajectory raise?
Trajectory raised $15M in Seed funding led by Conviction with participation from Bessemer Venture Partners, Radical Ventures, BoxGroup, Jeff Dean, and Fei-Fei Li.
Who founded Trajectory?
Trajectory was founded by Ronak Malde, Michael Elabd, and Arjun Karanam.
What does Trajectory’s platform do?
Trajectory provides continual learning infrastructure that captures corrections, retries, traces, edits, and re-prompts to help AI systems improve after deployment.
Why does continual learning matter for enterprise AI?
Continual learning allows enterprise AI systems to adapt from real-world usage instead of relying entirely on static pre-trained models, improving long-term performance and reliability.
What industries could benefit from continual learning infrastructure?
Industries deploying enterprise AI systems including software, finance, healthcare, legal technology, and cybersecurity could benefit from continual learning infrastructure.
What is Trajectory’s reported valuation?
Trajectory reportedly raised its Seed round at a $115M post-money valuation.
