On July 2, 2026, Microsoft announced the creation of Microsoft Frontier Company, a new operating business that will embed 6,000 industry and engineering experts directly inside customer organizations. The investment is $2.5 billion. Rodrigo Kede Lima, who spent six years leading enterprise transformations across the Americas and Asia, will serve as President. The unit will co-design, deploy, and continuously improve AI systems at client sites based on measurable business outcomes.
Microsoft did not frame this as consulting. It called Frontier Company something that "goes beyond what has been labeled as Forward Deployed Engineering" and described it as "the largest, most capable, outcome-driven engineering organization in the industry." The distinction matters to Microsoft. It should not matter to anyone else. When you send 6,000 employees to sit in other companies' offices and make their technology work, you are a consulting firm.
Two days earlier, AWS announced a $1 billion Forward Deployed Engineering organization. Engineers embed with clients in pods of five or six, running 45-day cycles. Teams are already deployed at the Allen Institute, Cox Automotive, the NBA, the NFL, Ricoh, and Southwest Airlines. On May 11, OpenAI launched The Deployment Company, a majority-owned subsidiary backed by more than $4 billion from 19 investment firms led by TPG, Advent, Bain Capital, and Brookfield. It acquired Tomoro, a London-based applied AI firm with approximately 150 forward deployed engineers. A week earlier, Anthropic formed an AI-native enterprise services firm with Blackstone, Hellman & Friedman, and Goldman Sachs as founding partners.
Four platform providers. Sixty days. More than $9 billion committed. Tens of thousands of engineers. All to solve a problem their platforms were supposed to solve on their own.
The model they are all copying was invented by Palantir around 2010, when intelligence agencies could not articulate requirements through normal procurement channels. Palantir embedded engineers at client sites who could observe the actual workflow, build on the spot, and iterate daily. It was considered niche, unscalable, and specific to government. Between September 2020 and mid-2025, Palantir stock returned approximately 640 percent. The niche model produced the return. Now every major AI company is replicating it.
The reason is legible in the data. MIT's NANDA Initiative studied 300 public AI deployments and found that 95 percent of enterprise AI pilots produced little or no measurable impact on profit and loss. The problem was not the models. It was the deployment. The models worked in demos. They failed in production. The gap between a working prototype and a working enterprise system turned out to be filled not with software, but with people.
Microsoft's own product illustrates the gap. Microsoft 365 Copilot crossed 20 million paid seats in April 2026, but that represents only 3.3 percent of its commercial user base. Of those with access, only 35.8 percent actively use it. The product's Net Promoter Score hit negative 24.1 in September 2025 and recovered only to negative 19.8 by January 2026. Some 44 percent of lapsed users cited distrust of answers as their reason for stopping. Microsoft's paid AI subscriber share fell 39 percent between July 2025 and January 2026, dropping to 11.5 percent behind ChatGPT at 55.2 percent and Gemini at 15.7 percent.
Microsoft stock is down 21 percent year to date, the worst performance among mega-cap technology companies. Frontier Company is the response. Not a better model. Not a better interface. Six thousand people in your office, making the existing product work.
The competitive irony runs deeper. The US AI consulting market exceeds $15 billion in 2026. Accenture generated $2.7 billion in AI revenue and $5.9 billion in generative AI bookings in fiscal 2025. Deloitte committed $3 billion to its own generative AI investment. These are the firms that AI was supposed to displace. Instead, the AI companies are entering their market, competing for the same contracts, hiring the same people, doing the same work. Accenture and ServiceNow have even launched their own forward deployed engineering program, copying the AI companies that are copying Palantir. The consultants are imitating the disruptors who are imitating them.
The connection to last week's cost crisis is direct. Tesla capped employee AI spending at $200 per week. Uber burned its 2026 AI budget by April. Microsoft's own Experiences and Devices division canceled Claude Code licenses by June 30. These companies discovered that AI tools have marginal costs that compound when thousands of engineers use them daily. Microsoft Frontier Company is the other side of the same coin: it is not enough to buy the tools. Someone has to make them work. And the company that sells the tools just committed $2.5 billion to be that someone.
The entire trajectory of enterprise software from 2000 to 2024 pointed in one direction: fewer people, more platform, self-serve, scale. Cloud computing eliminated on-premises engineers. SaaS eliminated implementation consultants. Every generation of software promised to remove one more human from the process. AI was supposed to be the final step, the technology that made everything else self-serve.
Instead it reversed the direction. The technology that was supposed to replace human expertise now requires more of it. The platform that was supposed to be self-serve requires 6,000 employees to install. The product is not the model. The product is the house call.