Funding Brief｜Trajectory Series A $40M

🌏 中文版

Funding Details

What the Company Does

Trajectory builds continual learning infrastructure — enabling deployed AI Agents to keep getting smarter from real-world signals like user corrections, re-prompts, and edits, rather than staying frozen at whatever capability they had at training time.

The core product is a lightweight SDK that, once integrated into a customer's product, captures user traces, corrections, re-prompts, and edits. It then uses "self-distillation policy optimization" to programmatically update the Agent's future decision paths based on these real-world corrections — regardless of whether the customer runs open-source or closed-source models. Optimization can happen at the prompt level, model parameters, or even the entire agent harness, without relying on a consulting team to hand-tune each deployment.

Current customers include Clay (AI sales), Decagon (AI customer service), Harvey (AI legal), Mercor, and Rogo — all well-known AI-native companies, some already in production. The founding team comes from Google DeepMind, Apple, OpenAI, and Meta Superintelligence Labs. The company emerged from stealth in May 2026.

What This Round Signals

Implications for the Agent Ecosystem

The key signal from this round: the competitive focus in the Agent space is shifting from "who has the stronger base model" to "who can make deployed Agents learn on their own." Previously, enterprises wanting to customize open-source models had to rely on consulting teams for manual fine-tuning. Trajectory automates this process, optimizing directly at the agent harness level — how the Agent calls tools, how it handles errors — rather than simply swapping in a larger base model.

What Investors Are Betting On

Sequoia went from sitting out the Seed round to leading the Series A just 3 months later (Bessemer is the only investor that participated in both rounds). The bet is that continual learning will become a required middleware layer for all agentic products — every company deploying models needs a continuous learning loop instead of retraining from scratch each time. NVIDIA joining as a participant follows its consistent strategy of positioning itself at critical infrastructure nodes for startups that consume heavy compute.

Numbers Worth Watching

• Valuation jumped from $115M (Seed) to $300M (Series A), roughly 2.6x growth in 3 months — an unusually fast fundraising pace;
• A 14-person team has raised $55M in total, over $3.9M per person — reflecting the premium placed on elite research backgrounds (DeepMind, OpenAI, Apple, Meta Superintelligence) rather than proven revenue;
• The customer roster (Clay, Decagon, Harvey, Mercor, Rogo) consists entirely of multi-billion-dollar agentic startups, showing Trajectory's "sell picks and shovels to the gold rushers" strategy — positioning itself as core infrastructure for other AI unicorns.

Watchlist Status

Trajectory is not yet on the watchlist. Recommended for section B6 (Agent Observability/Evaluation), tracked alongside Braintrust, Arize AI, and LangSmith. Key tracking angle: continual learning infrastructure that turns real-world usage signals (traces/corrections) into ongoing agent decision-path optimization. $40M Series A, $300M valuation.

Takeaway

The conventional wisdom was that "Agent optimization" mainly meant swapping in better base models or writing more refined prompts. Trajectory's approach is a reminder that the truly hard-to-replicate optimization signals are buried in production correction logs — whoever can automatically turn those signals into training data holds a more durable advantage than simply upgrading models.

References

• Trajectory raises $40M Series A at $300M valuation — Dealroom
• Trajectory Raises $40M in Series A Funding at $300M Post-Money Valuation — FinSMEs
• Trajectory Raises $40M Series A — The SaaS News