Microsoft Frontier Firm Playbook: Human Value Keeps Moving

Why AI's Generic Output Misses Your Company's Secret Sauce

Ask any AI tool to write a sales pitch, analyze a market or draft a strategy memo and you will get competent, generic output, a convincing average of everything the model has seen. That may be fine for a first draft. It is dangerous for the things that make your company distinctive.

Your sales approach is not the average of every sales approach. Your most experienced people know which trade-offs matter, which exceptions deserve attention and which deals are likely to close. That is specific, hard-won knowledge, and much of it is never documented. It lives in people’s heads.

Every organization runs on two kinds of knowledge: explicit knowledge, documented in processes, systems and databases, and tacit knowledge, the judgment, instinct and institutional memory of how work really gets done. Microsoft’s Becoming a Frontier Firm playbook, developed from more than 100 internal AI transformation efforts, calls this the organization’s “secret sauce.”

The Four Dimensions of Organizational Advantage and Private Evals

Every organization runs on two kinds of knowledge: explicit knowledge documented in processes, systems and databases, and tacit knowledge — judgment, instinct, institutional memory and the unwritten understanding of how work really gets done. Microsoft identifies four dimensions of this organizational advantage: the company’s point of view, its proprietary performance knowledge and institutional judgment, its definition of what good looks like, and the guardrails that determine where AI should act, where humans should review and where AI should not be deployed at all.

Microsoft calls the mechanism for capturing this “private evals”: company-specific standards that translate tacit knowledge, judgment and definitions of quality into criteria AI systems can be evaluated against. Rather than allowing the system to converge toward a generic definition of quality, the organization continually tests it against its own definitions.

That is strategically important because, as access to powerful AI becomes more widespread, the models themselves become less differentiating. Competitive advantage increasingly comes from what the organization can teach those models about how it creates value — shifting the competitive question from who has access to the best AI to who has built the better learning system around it.

How AI Learns Tacit Knowledge: The Hill-Climbing Machine

It is easy to read this through the familiar lens of AI replacing human expertise. Katy George, Microsoft’s corporate vice president of AI and Work Transformation and a co-author of the playbook, described something more dynamic: humans and AI together achieve new capabilities, new things that weren’t part of the human-only process.

The playbook calls this a “hill-climbing machine,” a learning system where an organization defines what good looks like, supplies the context and uses feedback to improve the system over time. With every cycle of feedback, scoring and tuning, the AI becomes more aligned with the organization’s own standards rather than converging on generic outputs.

As the system gets better at applying what the organization already knows, people gain new capacity to test ideas, solve different problems and redefine what good looks like next. Competitive advantage therefore starts shifting from what the organization knows to the speed of organizational learning: how quickly it can create, distribute and renew what it knows.

Human Value Keeps Moving: New Skills, T-Shaped Roles and the Expertise Pipeline

As AI makes accumulated knowledge available to more people, expertise cannot derive only from having knowledge others do not. Its value increasingly comes from extending it: noticing when yesterday’s rule no longer applies, recognizing a pattern nobody has coded for yet, questioning what good looks like and seeing an opportunity the data does not yet show.

The playbook emphasizes judgment and taste, business acumen, learning agility and the ability to scope and delegate work to agents. Roles become more T-shaped, with deep disciplinary expertise still important but greater expectation that people connect work across the value chain; engineers may operate more like full-stack product builders. That requires more fluid operating models, including end-to-end value streams and dynamic teams.

Organizations must also keep producing people capable of generating new judgment. Microsoft’s PRAISE program pairs early-career engineers with experienced mentors in a two-way learning model: the senior teaches craft while the junior brings AI fluency. As George says, humans will aim, train and govern the machine—but the human operating model and org structure that support that remain unresolved.