The standard enterprise AI playbook — sell a subscription, provide an API, and leave the integration to the customer — is hitting a wall. Companies are finding that moving beyond experimental chatbots into production-grade workflows is significantly harder than anticipated. In response, the industry is seeing an early signal of a shift: tech giants are moving away from selling software licenses toward selling embedded expertise.
Microsoft launched its Frontier Company on July 2, 2026, backed by a $2.5 billion investment and 6,000 engineers. Judson Althoff, CEO of Microsoft Commercial Business, described the unit as “the largest, most capable, outcome-driven engineering organization in the industry.” Rather than simply shipping code, these engineers are being embedded directly at customer sites. At 3M, for instance, Microsoft engineers are working within the Global Business Services team to build custom AI agent workflows for tasks like credit checks and delinquency assessments. Crucially, this platform is model-agnostic, integrating AI from Microsoft, OpenAI, Anthropic, Google, and open-source providers. To ensure reliability, the system includes human-in-the-loop controls and a custom monitoring dashboard that gives 3M staff real-time visibility and approval capability. This is a services-heavy model, not a traditional SaaS play.
This strategy is not unique to Microsoft. In June 2026, AWS launched its own $1 billion Forward Deployed Engineering (FDE) unit, which utilizes a nearly identical model of placing engineers inside client organizations. Analysts observing these moves suggest the competitive battleground is shifting away from who has the most powerful model and toward who can successfully own the “last mile” of integration.
However, this high-touch approach faces significant skepticism. Critics argue that this is essentially rebranded consulting, similar to the models used by firms like Accenture or EY. From a financial perspective, this is a departure from the high-margin world of software licensing. Services revenue is notoriously harder to scale and carries lower margins than pure SaaS, raising questions about how sustainable this model will be for tech giants accustomed to software-level profitability.
There is also the issue of trust. Patrick Moorhead, CEO of Moor Insights & Strategy, noted via Reuters on July 2 that large businesses “suspect that using models from Anthropic and OpenAI will eventually grant those frontier labs expertise to compete with them, especially in fields such as coding and law.” By embedding engineers, Microsoft and AWS are attempting to bridge that trust gap, but the underlying data-foundation issues — governance, security, and compliance — remain significant hurdles.
The stakes are high because the current track record for enterprise AI is poor. According to MIT’s Project NANDA, roughly 95% of enterprise generative-AI pilots in 2025 failed to deliver measurable business impact. The difficulty of integrating AI into production workflows is the primary culprit, and it is unclear if simply adding more human engineers will solve the fundamental complexity of organizational change.
While Microsoft’s work with clients like 3M, Unilever, and Novo Nordisk shows a clear intent to solve these integration problems, it is too early to call this a fundamental shift. It is an expensive, labor-intensive experiment designed to force AI into production. Whether this “last mile” strategy can overcome the high failure rates of previous AI pilots — or if it will simply become a high-cost consulting arm — remains the central question for CIOs watching this space.