Moat in Proprietary Workflows

The moat moves from the wrapper to the ground truth. If every lab can plug into the same gym style interface, the scarce asset is not the API layer, it is the underlying task set that mirrors real work, the system access needed to recreate that work, and the verifier that can tell whether an agent truly succeeded or just found a shortcut. That is why Fleet’s edge depends on realistic cross system environments and fresh, contamination resistant evaluations.

• Exclusive tasks matter because real enterprise work usually crosses several systems. Fleet already models workflows like claims processing and CRM updates across changing state, where a native app owner like ServiceNow is strongest only inside its own stack.
  1 sacra 4 servicenow 5 servicenow
• Proprietary workflow access matters because the hardest training data is locked inside customer software and internal process logic. Fleet’s early revenue came from custom builds for financial services and insurance, where each engagement leaves behind reusable task libraries, failure cases, and domain specific verifiers.
  1 sacra
• Hard to replicate verifiers matter because labs are building more of the tooling around agents themselves. OpenAI offers graders, tool use, and connectors in its API, and Anthropic ships computer use in a sandboxed environment, so generic environment plumbing is easier to copy than evaluator logic tied to real workflows.
  1 sacra 3 openai 6 openai 7 openai 8 anthropic

This pushes the category toward owners of proprietary workflow reality. The winners will be the companies that can refresh tasks quickly, span many systems instead of one product surface, and prove that passing the benchmark predicts working correctly in production. That is the path for Fleet to stay upstream of labs, incumbents, and open source standards.