Two sets of data from Q1 2025 are particularly revealing when viewed side by side.
On one side, traditional software engineering positions have plummeted 70%. On the other, demand for "Forward Deployed Engineers" (FDE) has accelerated from an 800% growth rate straight to 1000%. This decline and surge aren't just about numbers — they represent a fundamental redefinition of what "engineer" means across the entire AI industry.
OpenAI clearly saw this inflection point earlier than most.
$4 Billion Buys More Than Money — It Buys "People"
On May 11, OpenAI officially announced the formation of the OpenAI Deployment Company, with an initial investment exceeding $4 billion, partnering with 19 institutions including TPG, Bain Capital, Advent, Brookfield, Goldman Sachs, and SoftBank. But that's not the headline — what truly matters is the move they made simultaneously: the acquisition of Tomoro.
Tomoro is an AI consulting firm founded in 2023. Not large, but sharp. Its core business isn't training models — it's helping enterprises actually integrate OpenAI's models into their business systems: data ingestion, access control, production-grade workflow design. It handles all the gritty, unglamorous work. Its client roster includes Mattel, Red Bull, Tesco, Virgin Atlantic, and Supercell — none of them "tech companies" in the traditional sense.
What OpenAI wants is the 150 on-site deployment engineers Tomoro has on its roster. After the acquisition, these people become OpenAI's standing army, ready to deploy directly to customer sites.
It's a pragmatic play. No matter how powerful the model, connecting to a customer's on-site systems is connecting to their on-site systems — JPMorgan's data structures, compliance requirements, internal politics, and what problems they actually want to solve. That knowledge doesn't live in OpenAI's Mountain View offices. It lives in the customer's server rooms and conference rooms.
What Exactly Is Tomoro?
Tomoro has been tagged with the "OpenAI ecosystem" label since day one. What it does, in simple terms: put your engineers sitting next to the customer's engineers, and get AI running together.
The standard software delivery pipeline goes: build a product → sell it to a customer → let the customer figure out how to use it. But enterprise customers' real environments are always "unique and complex" — legacy systems, regulatory constraints, internal processes that were never designed with AI in mind. SaaS products hit a wall here.
The FDE model flips this chain: model companies send their best engineers directly into customer organizations, sitting alongside engineers who understand the customer's business, shipping real code, building custom integrations. Two knowledge systems colliding in the same space — project success rates improve dramatically.
Notably, Tomoro's homepage features this line:
"Our mission is to balance the productivity of artificial intelligence with human goals, making a three-day work week a reality."
An AI deployment company, putting a three-day work week on its homepage. It's not a gimmick — they're genuinely hiring on-site engineers in Australia, Singapore, and the UK.
Anthropic's Counter: BlackRock + Goldman Sachs, $1.5 Billion
OpenAI isn't the only company that sees this.
Last week, Anthropic announced a joint venture focused on enterprise AI deployment, valued at $1.5 billion. Founding partners: BlackRock, Hellman & Friedman, and Goldman Sachs. The three parties contributed $300 million together, with other investors including Apollo, General Atlantic, Singapore's GIC, and Leonard Green.
On one side, OpenAI + 19 PE firms, $4 billion and up. On the other, Anthropic + BlackRock + Goldman Sachs, $1.5 billion entering the arena. The timing is nearly synchronous — this is no coincidence. It's an arms race.
Over the past year, Anthropic has built strong momentum among developers and enterprise customers with its Claude lineup. OpenAI has internally acknowledged that Anthropic's growth constitutes "significant pressure," with head of applications Fidji Simo calling it a "wake-up call" in an all-hands meeting.
In a sense, the OpenAI Deployment Company is the product of that wake-up call.
From "Model Wars" to "Deployment Wars"
The shift in this competition is clear:
AI companies used to compete on who had more model parameters, higher benchmark scores, bigger context windows. Now the contest has become — who can embed their model into real enterprise business operations fastest.
The key success metrics for projects are also changing. Sixty to 70 percent hinges on "application deployment," not pure coding ability. Adaptability, leadership, soft skills — qualities that used to rank near the bottom of engineering hiring criteria — have now become the deciding factors.
OpenAI's head of platform engineering Sherwin Wu and product lead Olivier Godement described an extreme scenario on a podcast: in a physically isolated "air-gapped" environment at a national laboratory, deployment engineers must surrender all electronic devices, import model weights via physical media into supercomputers, and then perform "purely manual" environment adaptation for specific hardware.
This isn't writing code anymore. This is going to war.
Signals to Watch
For OpenAI's playbook to succeed, there are several key indicators worth tracking:
Customer conversion speed at the deployment company — $4 billion is no small sum. A target of 1,200 enterprises sounds aggressive, but the industry distribution and deployment velocity of the first batch of customers will determine whether this model is replicable.
The FDE talent supply bottleneck — People who understand both AI and enterprise system architecture are already scarce. Tomoro's 150 engineers are the first wave, but scaling requires a talent pool far larger than that. Anthropic will be fighting for the same people.
Can the three-day work week actually be delivered? — This isn't a feel-good issue. If AI deployment engineers can genuinely accomplish in 3 days what takes others 5, that alone becomes the best advertisement for productivity in this industry.
The next major battleground in AI competition may not be on benchmark leaderboards, but in customer conference rooms and server rooms. Whoever can embed their engineers fastest wins.
Primary sources: InfoQ (2026-05-12), OpenAI official statement, Tomoro website, Reuters reporting