Frontier Labs are investing into Forward Deployment because AI doesn't work on its own.
For two years, the pitch to boards was that software was autonomous. Buy an API key, stream tokens, and your business transforms overnight. Now the biggest frontier labs in the world are spending over on consulting consortia and armies of forward-deployed engineers. When the software cannot run without a human walking the floor, the software does not work on its own.

- Anthropic committed to the Claude Partner Network to subsidize consultancies and fund dedicated Applied AI engineers on live enterprise deals. It backed that with a .5bn services venture alongside Blackstone and Goldman Sachs to put engineers inside customer offices.
- OpenAI closed a deployment company with TPG, Brookfield and Bain Capital. The target is companies that bought ChatGPT Enterprise seats and got nothing out of them without senior engineers sitting at their desks.
- Google Cloud and Accenture stood up the Gemini Enterprise group with 1,000 forward-deployed engineers on top of 50,000 certified staff. Accenture's lead Chetna Sehgal put the problem in plain words: enterprises need true value from AI, and they are stuck.
- Raw models calculate probabilities. They do not know your ERP, they do not understand procurement indemnities, and they carry no operational conscience. Human intervention is the permanent bridge between raw intelligence and business value.
For two years, the labs sold Silicon Valley's favorite dream: autonomous enterprise software. The pitch to every board was simple. You sign the contract, you plug in the model, and the machine automates your operations while you sleep. Enterprise CIOs spent millions on token quotas, seat licenses, and API credits.
Then reality caught up. The tokens arrived. The business value did not.
A raw language model has no conscience. It does not know your ERP schema. It does not understand why your supply contracts have custom rebate structures. It cannot take responsibility when a billing pipeline drops millions of dollars on a weekend. A model calculates probability. It does not carry operational responsibility.
Now the balance sheets are confessing what the pitch decks denied. Over in capital has shifted from training models to paying for human labor.
On May 4, Anthropic and OpenAI launched parallel private-equity backed services ventures within hours of each other. Anthropic raised .5bn with Blackstone, Apollo and Goldman Sachs. OpenAI raised with TPG, Brookfield and Bain Capital. Both ventures exist for one reason: to embed forward-deployed engineers inside customer offices because enterprise clients cannot get value out of raw models on their own.
Anthropic doubled down by pouring into its Claude Partner Network. It is paying consultancies, subsidizing technical certifications, and sending dedicated Applied AI engineers to sit on live customer deals.
Google followed the exact same playbook. It partnered with Accenture to build a 1,000-person forward-deployed engineer army, backed by 50,000 Google Cloud staff. Chetna Sehgal, Accenture's Gemini Enterprise lead, admitted what every operator already knows: enterprises are stuck. They bought the technology, but they cannot connect it to a real business result without humans in the room.
This is the rise of human intervention. It turns out that making intelligence useful requires taste, domain expertise, and operational judgment. You cannot prompt your way around twenty years of institutional knowledge. You need a human being who knows where the money leaks and how the systems talk to each other.
The danger for enterprise buyers is how this human intervention gets billed.
The legacy systems integrators want to sell you a permanent payroll line. Accenture will gladly rent you 1,000 bodies at an hour, billing you month after month to babysit a model you do not own. When the consultants pack up their laptops, your enterprise is left with nothing but an ongoing invoice.
The real answer is owned agentic systems. You bring in senior forward-deployed builders to wire the models directly into your operations, build the safeguards, and leave behind an owned capability that your team runs on your balance sheet.
The labs spent billions trying to replace humans. They just spent billions more hiring them back.
Let's talk AI


