Matt Fitzpatrick: Who Wins the Data Labelling Race & Why Al Needs Forward-Deployed Engineers
Matt Fitzpatrick of Invisible Technologies argues enterprise AI adoption requires forward-deployed engineers, not off-the-shelf agents, and that synthetic data will not replace human feedback.
- MIT report: only 5% of enterprise GenAI deployments are working in any form; Gartner predicts 40% of projects canceled by 2027.
- Externally driven AI builds are 2x as effective as internal team builds per the same MIT report.
- An e-commerce retailer spent $25M building a returns agent, built their own eval tool, then shut it down months later and reverted to deterministic flow.
- Invisible offers 8-week free proof-of-concept sprints and charges zero for forward-deployed engineers — competitors charge for FDEs.
- The synthetic data will replace human feedback claim is the biggest misnomer in the industry; Fitzpatrick says it fails basic first-principles reasoning.
- Out-of-the-box enterprise agents test at ~58% accuracy on single-turn and ~33% on multi-turn workflows, meaning they largely do not work.
- Invisible has 1.3 million domain experts per year flowing through its workflows across a five-module platform (Neuron, Axon, Atomic, Meridian, Synapse).
- Fitzpatrick is most excited about AI in education: cheap personalized learning could disrupt a broken college/debt system and close STEM gaps by socioeconomic status.
2025-12-31 · Watch on YouTube
