The $10 Billion Bridge
MIT says 95% of enterprise AI pilots fail. The AI labs' answer, tallied up last week: roughly $10 billion committed to putting humans on-site to fix that. OpenAI's $4 billion services venture led by TPG, with a guaranteed 17.5% minimum annual return per FT reporting cited on the show. Microsoft's $2.5 billion Frontier division. A $1 billion org inside AWS. $750 million from Google. Anthropic's services venture with Blackstone and Hellman & Friedman. The forward-deployed engineer, a job Palantir invented in 2005, is suddenly the hottest business model in software, and the market read it exactly that way: "the market treated this announcement by OpenAI getting into the FDE and consulting business as an extinction event for professional service firms." India's IT index dropped 3.7% on the news.
The why is not complicated. "AI pilots fail because they're pilots," Nate B. Jones argued this week: label something a pilot and it gets fewer resources, de-risked goals, and no path to production. His estimate is that enterprise AI is "about 20% of technology, probably, and 80% of people problem," and his read on the lab programs is blunt: "you can't just stick a raw LLM into an enterprise and expect magic to result. You actually have to put engineers on the ground."
Recall last week's column: Decagon called FDE-as-strategy a trap that turns you into a glorified consultancy. The rebuttal didn't take long, but the trap argument showed up inside the FDE episode itself: $300-800 an hour as a permanent line item, custom code nobody can maintain, labs harvesting your workflows into vertical products, and "companies are basically running out of budget before they run out of enthusiasm."
The productized alternative got its own hearing. Asana's product chief Arnab Bose cited BCG data that ~75% of knowledge workers use AI while only ~5% of companies see meaningful gains, because "Those workflows are not actually being accelerated by AI as yet." His pitch is agents as governed teammates, with profile pages, human managers, and audit trails, software doing what FDEs do by hand.
And the week's best kicker came from Tokyo. Bespoke founder Akemi Tsunagawa spent years faking her tourism chatbot before the AI caught up: "We also had human, including myself, chat with people, pretending to be a bot." The bridge between AI promise and AI production has always been humans. The only thing that changed this year is who sends the invoice.
Sources: Ray Rike (Benchmarkit) and Peter Buchanan (New Plan), Metrics that Measure Up, "Forward-Deployed Engineers (FDEs) - AI's New ROI Battleground," Aug 4, 2026. Nate B. Jones, CXOTalk, "Why Your Enterprise AI Pilot Won't Scale," Aug 6, 2026. Arnab Bose (Asana), Dev Interrupted, "Why AI gains are unevenly distributed in your engineering team," Aug 4, 2026. Akemi Tsunagawa (Bespoke), This Week in Startups, "How Bespoke faked AI until it actually worked," Aug 3, 2026.
Enterprise AI is 80% a people problem; labs are selling humans as the bridge while Asana productizes the same work as governed agent teammates.