Stop sending me engineers: what OpenAI, Anthropic and Microsoft misread about enterprise AI
MIT found 95% of enterprise AI pilots produced zero measurable profit. The labs' answer: send more engineers. The real problem: you need to send fewer.
Why it matters
Enterprise AI's forward-deployed engineer model treats a delivery problem as a diffusion problem, outsourcing business knowledge acquisition rather than building internal AI capability. The structural misalignment—vendor engineers optimized for quarterly conversion metrics vs. enterprise buyers needing multi-quarter, cross-functional value realization—means scaling AI into the wider economy requires inverting the talent flow: deploying domain experts into AI literacy, not AI experts into business domains.
The key facts
9 to knowMIT study: ~95% of enterprise AI pilots produced no measurable profit impact despite tens of billions spent
OpenAI Deployment Company: $4B initial investment, 150 FDEs at launch (May 2026)
Anthropic: enterprise AI services venture with Blackstone, Hellman & Friedman, Goldman Sachs, reportedly >$1.5B valuation
Microsoft Frontier Company: $2.5B, ~6,000 embedded engineers and industry experts (July 2026)
AWS: $1B commitment to forward-deployed engineering unit
FDE job postings grew ~800% in nine months; principal-level compensation exceeds $1.2M
Constellation Research flagged: FDEs often used as crutch for product immaturity and vendor dependency play
Dice reported August 2026: shortage of engineers directly holding back enterprise AI returns, scarcest talent concentrated in Fortune 50 accounts
Author's case study: internal redesign of supply-chain planning architecture using domain operators + AI achieved persistent inventory reduction vs. external model deployment
Go to the source
CIOcio.com
Publisher excerpt: The AI industry says enterprises can’t absorb AI. Its answer is to absorb it for them. That’s not the same thing. The engineer across the table knew the model cold. Not vendor-deck cold. He could tell me where it would break, what it would cost, and which benchmark numbers to ignore. One of the…