The Information reports that Meta is launching a new enterprise solutions division that will embed engineers and product managers within enterprise customers and drive their use of the company’s AI tools. The division, led by product director Naomi Gleit, is the latest sign that the most expensive jobs in enterprise AI are no longer researchers building frontier models, but engineers diving into customer sites to make those models work.
The move makes Meta the latest major AI company in a 13-week series of deployment-focused enterprise announcements. On February 23, OpenAI announced the Frontier Alliance, a partnership between its Forward Deployed Engineering team and McKinsey, Boston Consulting Group, Accenture, and Capgemini. On May 4, Anthropic announced a joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs that would combine the company’s engineering resources into a standalone enterprise services company, valued at more than $1.5 billion. A week later, OpenAI announced The OpenAI Deployment Company, capitalized at more than $4 billion and led by TPG with Advent, Bain Capital and Brookfield as co-lead founding partners, and acquired Edinburgh-based Tomor for approximately 150 deployment engineers. Together, the two ventures will inject approximately $5.5 billion into AI deployment and enterprise services.
A Forward Deployed Engineer (FDE) is a senior software engineer who resides within a customer’s environment, learning its data and politics, and shipping production code against the disruption caused by legacy IT systems. Palantir pioneered this title more than a decade ago, calling early embedded engineers “deltas.” This model has long looked like a margin killer to traditional software investors, but most of the industry ignored it until adoption became a binding constraint. Palantir’s Q1 2026 results, with sales up 85% year-over-year and 133% growth in its U.S. commercial business, gave the market new evidence of a model already gaining momentum.
Why Model Access stopped closing deals
That’s changing because access to raw models is no longer enough to win an enterprise account. A 2025 MIT report found that about 95 percent of enterprise-generated AI pilots showed no measurable P&L impact, and researchers pinned this failure on flawed integration rather than weak models. The bottleneck was not the functionality of the model, but the gap between sophisticated demonstrations and working integration within legacy systems, a change advisory board and compliance structure that moved at its own pace.
OpenAI executives have been upfront about this. Business customers already account for more than 40% of OpenAI’s revenue, and the company expects to match consumer revenue by the end of the year, CNBC quoted chief revenue officer Dennis Dresser as saying. This combination only holds true if the pilot converts to production, and the Institute concludes that conversion will not occur without engineers in the field.
Anthropic has been down this path for some time. The company deployed Claude to more than 470,000 Deloitte employees in October 2025, in what remains the company’s largest enterprise deployment. Then, in December, we expanded our Snowflake partnership to a multi-year, $200 million deal that will bring Cloud to more than 12,600 Snowflake customers across AWS Bedrock, Google Vertex AI, and Microsoft Azure. Hyperscalers are not bystanders, so cloud-to-cloud details matter. They own the data plane where the workloads run, manage the procurement, and in many accounts distribute the models that the adopting company must handle.
Conflict with SIer
The strategic question for buyers is who will actually do the hands-on work. Before 2026, most of it flowed through global consulting firms, offshore-led integrators, hyperscalar professional services, and specialty boutiques, with Accenture, Deloitte, TCS, Infosys, and Capgemini positioning their AI service lines to absorb the next wave.
Its position is currently being contested. Anthropic’s joint venture is positioned in the announcement as a way to expand its supply of skilled implementation partners, and Anthropic is careful to call it an addition to, rather than a replacement for, existing integrator relationships. Still, by placing Claude across portfolio companies such as Blackstone, Hellman & Friedman and other supporting investors, the new company creates a parallel channel that can bypass traditional integrators for these accounts.
OpenAI adoption companies are an even more immediate challenge. The company is a majority-owned subsidiary of OpenAI that hires engineers, bills customers, and makes acquisitions. OpenAI also operates the Frontier Alliances program with four major consultancies. If you read this honestly, while McKinsey, BCG, Accenture, and Capgemini have certified OpenAI technology practice groups, OpenAI is building an in-house alternative, so the same vendors are both partners and competitors.
System integrators in India are facing the toughest situation. While Infosys, TCS, Wipro and HCLTech have competed on cost-effective supply of engineering talent at scale, lab-owned companies value proximity to models and direct access to roadmaps. Forward deployment engineers at OpenAI implementation companies can route customer requests directly to the product in a way that third-party engineers cannot.
What the Institute cannot do yet
The economics of the FDE model has not been proven outside of Palantir. Industry trackers report sharp increases in postings over the past year, but numbers vary widely by methodology, and senior technical roles at OpenAI and Anthropic can reach high six figures, although it’s unclear whether FDE salaries are that high across thousands of implementations. These numbers only work if the lab can produce reproducible patterns from each effort. Palantir spent 10 years building Foundry and AIP to transform artisanal FDE work into reusable platform IP. Frontier Labs is betting it can do similar work much faster.
Coverage is another open question. Lab-owned companies start with strong accounts, private equity portfolio companies and existing corporate relationships, rather than mid-market or regional clients where Indian integrators have built deep practice areas. Things get even more complicated as hyperscalers’ consulting arms compete with and partner with new companies like AWS, Google Cloud, and Microsoft.
What CXOs should take away from this
For technology buyers, the implications are tangible. The implementation market is divided into broad stages rather than clear categories. The best companies are those that own labs where engineers have direct access to model behavior and roadmaps. A global consultancy sits somewhere in between, with multi-platform certification and change management capabilities. Offshore-led integrators compete on scale, cost, and tenure. In reality, one program still incorporates model labs, hyperscalers, data platform vendors, and often boutique shops all at once.
Questions about ownership are therefore of pressing value. CIOs negotiating new AI engagements should ask whether the implementation team has feedback channels to model owners, who will manage the architecture and data integration, and what happens if the underlying model is deprecated or the price changes. While the Institute is building a bunker out of its implementation workforce, an open question is whether buyers will eventually become locked into a single model family, just as companies were once locked into a single ERP vendor.
Meta entries are a useful signal. If a company with perhaps the strongest open weight story in the industry decides that selling Llama through its partners isn’t enough, that conclusion is hard to avoid. Access to models has become a commodity input, with the labs that benefit the most spending billions of dollars on engineers to install these models within customer environments, and roughly $5.5 billion in OpenAI and Anthropic ventures alone.
