The AI models already exist. The problem starts the moment you try to bolt one onto your company's systems.
That's why OpenAI, Anthropic, and Meta have poured $5.5 billion into this one problem recently. Specifically, into putting more people on it.
Everyone believed the same thing
Bolting a chatbot onto a Fortune 500 company is easy. Getting that AI to actually run inside a 20-year-old SAP instance or a heavily customized CRM is a completely different problem.
So the industry's answer became the Forward Deployed Engineer (FDE). Palantir pioneered the model a decade ago, when intelligence-agency clients couldn't articulate what they wanted — the fix was to embed senior engineers inside the client, shipping production code amid the client's own data, office politics, and legacy-system chaos.
That model has now spread across the entire AI industry. OpenAI, Anthropic, and Meta have collectively put about $5.5 billion into FDE-based deployment firms. FDE job postings are up 800% year-over-year as of 2026. At Palantir, average total comp for an FDE runs $238,000, with staff-level roles clearing $630,000. At OpenAI and Anthropic, mid-to-senior FDEs land in the $350,000–$550,000 range.
The logic is simple. Access to a model alone doesn't close an enterprise deal. Someone still has to get inside the client and make it actually work.
Benioff bet the opposite way
On August 3, a $20 million pre-seed round led by Marc Benioff's Time Ventures went public. The company: June, just out of stealth. Its CEO, Efrat Rapoport, previously sold conversational AI startup Bonobo to Salesforce for $45 million in 2019, alongside her three co-founders.
What June sells isn't headcount. It's a platform where an AI agent scans a company's systems first, extracts how the business actually runs, and auto-generates a deployment roadmap. Instead of people spending weeks interviewing stakeholders for requirements, the AI draws the map first.
The numbers explain why this matters. IDC found that of every 33 AI pilots companies launch, only 4 reach production — an 88% failure rate. MIT's GenAI Divide report is even blunter: 95% of pilots delivered no measurable financial return. And the report pinned the cause not on model quality, but on legacy system integration.
Rapoport's own framing cuts right to it. "AI, paradoxically, increases the demand for professional services. The hard part isn't the demo — it's changing existing systems and workflows."
There's already a real case. Paul Akinmade at mortgage lender CMG adopted Claude Code quickly, then got stuck at Salesforce integration. After using June, his team could see exactly where to deploy agents — and move forward safely without bringing in an FDE.
| Embedded FDE model | AI auto-roadmap model (June) | |
|---|---|---|
| Cost structure | $238K–$630K+ total comp per hire (Palantir baseline) | Platform subscription (pricing undisclosed) |
| Time to start | Months of hiring and onboarding | Roadmap right after a system scan (vendor claim) |
| Scalability | Limited to headcount | Theoretically parallel across projects |
| Track record | A decade of Palantir results | Just out of stealth, August 2026 — one public case |
That last row matters. June isn't a proven methodology yet — it's a bet. But when Benioff, Michael Dell, Aaron Levie, and George Kurtz all put money on the same idea at once, it says something about how unsolved this problem still is.
What to check if this is your company
- Diagnose where you're actually stuck
If a pilot has stalled, figure out whether it's a model problem or an integration problem first. The IDC and MIT data both point to the latter. - Ask about reusability before signing an FDE contract
Find out whether the integration logic built for this engagement carries over to the next one. If it doesn't, you're paying full price every time. - Test roadmap-automation tools on coverage first
Confirm the tool can actually scan the legacy systems you run (SAP, Salesforce, in-house ERP) — with your own data, not a demo. - Default to combining people and AI
This isn't an either/or choice. Let AI handle the initial scan and roadmap, and keep people for the final stretch where office politics are involved. - Compare total cost to completion, not headline salary
An FDE alone runs $238K–$630K+ in total comp, while automation tools are usually subscription-based. Compare the full cost through to deployment, not just the sticker price.
Go deeper
June AI's official stealth-exit announcement Founder interviews, investor list, and the full 4-step product process globenewswire.com
Why Forward Deployed Engineers got so expensive From Palantir's origins to how much OpenAI and Anthropic are betting forbes.com
Why 88% of AI pilots never reach production The IDC research, plus the case that it isn't IT's fault cio.com
MIT's GenAI Divide — what the surviving 5% did differently Full context behind the 95% failure number legal.io
How Bonobo sold to Salesforce for $45 million June's founding team's first exit story voicebot.ai
The complete 2026 guide to FDE pay and hiring trends Comparing compensation bands across Palantir, OpenAI, and Anthropic hashnode.com
