Fleet AI
RL-environment vendor building simulated "gyms" of enterprise software where AI agents practice safely; ~$60M ARR, ~$750M valuation talks [1][2].
1. Core Product / Service
Fleet (Fleet AI, founded 2024) builds reinforcement-learning "gyms" — high-fidelity simulated replicas of enterprise software (e.g., Salesforce, Excel) in which AI agents practice, fail safely, and improve before deployment [1]. It sells these environments to frontier model labs and large enterprises, positioned as "send agents to the gym before they run the office" [1].
The company is one of the specialized RL-environment vendors local research calls the "new three" (with mechanize and HUD), in contrast to generalist data-labeling incumbents like scale-ai [local].
2. Target Users & Pain Points
- Frontier AI labs wanting realistic, safe agent-training environments for enterprise workflows.
- Large enterprises pre-deployment testing agents against their software stacks.
Pain solved: agent failure in production is expensive and unsafe; Fleet provides a sandbox that approximates real enterprise systems for training/evaluation [1].
3. Competitive Landscape
| Vendor | Focus | Vs. Fleet |
|---|---|---|
| Fleet | Simulated enterprise-software gyms | Enterprise-replica fidelity; strong revenue ramp |
| mechanize | SWE environments + graders | Code-focused; Google deal talks |
| scale-ai | Data labeling → RL envs | Generalist incumbent |
Fleet's differentiation is fidelity to enterprise software (Salesforce/Excel replicas) rather than pure code-repo environments [1].
4. Unique Observations
- The revenue ramp is the signal. From ~$1M annualized (2025) to ~$60M (April 2026), Fleet's growth shows RL environments shifting from experimental to mission-critical post-training spend for labs [2].
- A second RL-env reprice in the same week. Fleet's ~$50M round at ~$750M valuation (Bain-led) tracks mechanize's $500M seed mark — the "environment" layer is being repriced as fast as any AI-infra segment [1][2].
- Safety as a wedge. "Fail in the gym, not production" reframes simulation from a training cost to a risk-management product for enterprise agents [1].
5. Financials / Funding
- Founded: 2024, US-based (New York); ~40 employees [1][2].
- Seed: $15M (Sequoia Capital, Menlo Ventures, SV Angel) [2].
- Series A: led by Craft Ventures; total funding $45M+ [1].
- Next round (in talks): ≥$50M at ~$750M valuation, led by Bain Capital Ventures [1].
- Revenue: ~$60M annualized (April 2026), up from ~$1M (2025) [2].