Two giants, one week, the same bet
In May 2026, within a single week, the two giants of the AI industry placed almost the same bet.
Anthropic founded a new enterprise AI services company together with Blackstone, Hellman & Friedman, and Goldman Sachs. OpenAI launched "The Deployment Company" — an initial investment of over four billion dollars, nineteen partners.
The companies that hold the best models on earth arrived at the same conclusion at the same moment:
Building a smarter model is no longer enough.
If the people sitting on the frontier models are pouring billions into deployment, that tells you where the bottleneck actually moved. And if you've been watching job boards, you already saw the signal before the press releases.
The signal: a job title surging hundreds of percent
The clearest symptom is one role whose hiring demand has surged by hundreds of percent year over year: the Forward Deployed Engineer (FDE).
It's tempting to file this under "new high-paying engineering job, learn the skills, chase the comp." That reading misses the point entirely.
The FDE is not a job. It is the visible edge of a structural shift. As AI commoditizes the act of building, value collapses onto a single point — and the FDE is simply the first role standing on it.
BUILD got industrialized
For most of software's history, "being able to build it" was the value. Scarcity lived in the implementation: the people who could turn an idea into a working system were rare, and they were paid for that rarity.
AI dissolved that scarcity. Generating a working prototype, wiring an integration, scaffolding a service — the marginal cost of building is collapsing toward zero. When everyone can build, building stops being a moat.
So the obvious question: if "being able to build" is no longer the scarce thing, what is?
The 95% don't die in the lab. They die in the field.
Here is the number that frames everything. In MIT NANDA's study, against an estimated $30–40 billion poured into generative AI, roughly 95% of organizations have produced no P&L impact.
Read that carefully. The models work. The demos are spectacular. And 95% of the money still produces nothing on the income statement.
Those projects don't die in the research lab where the model is built. They die at the customer's front line, where the model is supposed to land — in the gap between "this works in a notebook" and "this changed how the business runs." That gap has a name in this book: the Valley of Death.
The 95% is not a model-quality problem. It's a deployment problem.
What's actually scarce now: outcome deployment
This is the one phrase the book is built around — outcome deployment (成果実装):
The power to grasp first-hand information at the customer's front line, turn a bespoke solution into a reproducible asset, and convert it into an actual business outcome — rather than merely building AI.
Notice what this is not. It's not prompt engineering. It's not picking the right model. It's the unglamorous, high-leverage work of crossing the Valley of Death: understanding a specific business deeply enough to make the technology produce a result that shows up in the numbers.
The one input AI can't synthesize
Why can't this be automated away like building was?
Because it depends on discovery — first-hand information from the actual front line. The messy, contradictory, context-bound reality of how a specific business actually works. That is the one resource a model cannot generate from its training distribution, no matter how large it gets. You cannot synthesize the ground truth of a customer you've never sat with.
When building is free, the scarce input is the thing you can only get by being there.
Palantir already proved it — twenty years ago
None of this is speculative. The prototype of outcome deployment was invented two decades ago, by Palantir — forward-deployed engineers embedded at the customer's side, turning bespoke, on-the-ground problem-solving into reproducible product. The industry is now rediscovering, at scale, a structure Palantir has been running since before "generative AI" was a phrase.
The difference between burning cash on bespoke contract work and building a durable business is one thing: productization — whether the bespoke solution you built at the front line becomes a reusable asset, or evaporates the moment the engagement ends.
What this means if you build for a living
The leverage didn't disappear. It moved.
- It moved from "can you build it" to "can you find what is actually worth building."
- It moved from shipping features to turning a frontline insight into a reproducible asset.
- It moved from the lab to the last mile.
If your entire value proposition is "I can implement," AI is now competing with you on price, and winning. If your value is "I can stand at the front line, find the ground truth no model can synthesize, and turn it into an outcome that moves the business" — you just became the scarcest person in the room.
That is the Forward Deployed Shift.
Read the full open-source book
I wrote The Forward Deployed Shift as an open-source book (CC BY 4.0), available in full in both English and Japanese. It connects, from primary data and field structure, the whole line: the industrialization of BUILD, the Palantir-origin prototype of outcome deployment, why discovery can't be synthesized, productization as the dividing point, and the D&V methodology for actually crossing the Valley of Death.
📘 GitHub — Leading-AI-IO/the-forward-deployed-shift
👉 https://github.com/Leading-AI-IO/the-forward-deployed-shift
It's part of a connected ecosystem of open-source books. Two good next reads:
- The Palantir Impact — the origin of outcome deployment: https://github.com/Leading-AI-IO/palantir-ontology-strategy
- Depth & Velocity — the methodology (the "OS") underneath it: https://github.com/Leading-AI-IO/depth-and-velocity
Issues and PRs are welcome — especially counter-arguments, and fresh cases from enterprise AI deployment.
Written by Satoshi Yamauchi (山内 怜史) — AI Strategist & Business Designer. More writing: https://note.com/satoshi_yamauchi
Top comments (0)