Give an AI tool a sales pitch, market analysis or strategy memo to produce, and the result will usually be competent and generic. It offers a persuasive average of what the model has encountered. That can be useful for an initial draft. For the work that makes your company distinctive, it can be dangerous.
Your approach to sales is not an average of everyone else’s. Your product taste is not the midpoint of every product decision. Your most experienced people understand which trade-offs count, which exceptions need attention and which deals are likely to close. Their knowledge is specific and hard-won.
Much of it has never been written down. People carry it in their heads.
Microsoft recently published its Becoming a Frontier Firm playbook, drawing on more than 100 internal AI transformation efforts. It calls this knowledge the organization’s “secret sauce.” For AI transformation, the intriguing question is what happens when companies start teaching that knowledge to AI.
Microsoft introduced the Frontier Firm in its 2025 Work Trend Index as an organization built around on-demand intelligence and hybrid teams of people and AI agents. The model signaled a move beyond simply using AI tools toward redesigning how work, roles and organizations operate.
As the Frontier Firm teaches AI to absorb more of the tacit knowledge and judgment that gave yesterday’s experts their value, human value has to evolve. It keeps moving as people recognize, decide and create what the system has yet to learn.
The Company Knowledge AI Has Yet to Learn
Organizations rely on two kinds of knowledge. Explicit knowledge is recorded in processes, systems and databases. Tacit knowledge includes judgment, instinct, institutional memory and the unwritten understanding of how work actually happens.
Microsoft describes 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 specifying where AI should act, where people should review and where AI should not be used at all.
Here, the human contribution becomes economically interesting. AI models may become increasingly common, while the accumulated judgment within a particular organization remains distinctive.
Microsoft calls these “private evals”: standards specific to the company that turn tacit knowledge, judgment and definitions of quality into criteria for evaluating AI systems. The organization continually tests the system against its own standards, preventing it from settling on a generic definition of quality.
That matters strategically because broader access to powerful AI makes the models themselves less differentiating. Competitive advantage increasingly rests on what an organization can teach those models about how it creates value. The competitive question shifts from access to the best AI toward the quality of the learning system built around it.
That raises another question: As the judgment that made your best people valuable becomes easier to codify into a system, what happens to those people?
How Organizations Teach AI Their Tacit Knowledge
It is tempting to interpret that question 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, outlined a more dynamic process in an interview for The Future of Less Work podcast.
“This is not about, let’s understand what’s tacit today and just have AI do exactly that,” George says. “What’s happening is that humans and AI together are able to achieve new capabilities, new things that weren’t part of the human only process.”
Transferring today’s human expertise into AI is one stage in a continuing process. As the system becomes more capable, people use that capability to take the work further.
The playbook calls this a “hill-climbing machine,” a learning system in which an organization defines what good looks like, provides context and improves the system through feedback over time. Each cycle of feedback, scoring and tuning brings AI closer to the organization’s standards and further from generic outputs. As the system improves at applying existing organizational knowledge, people gain capacity to test ideas, tackle different problems and redefine what good should look like next.
The loop works like this: people teach the system what they know, the system distributes that knowledge at scale, and the resulting capability enables people to attempt something previously beyond their reach.
Competitive advantage begins to shift toward the pace of organizational learning: how quickly an organization can create, share and renew its knowledge.
Why Human Expertise Must Keep Evolving
This changes the way we view expertise. As AI gives more people access to accumulated knowledge, expertise can no longer draw its value solely from knowing what others do not. Extending that knowledge becomes increasingly valuable.
For experienced professionals using AI, the work increasingly involves spotting when yesterday’s rule no longer holds, recognizing a pattern nobody has coded for, questioning what good looks like and identifying an opportunity the data has yet to reveal. That asks far more of people than knowing the answer.
The implications are already visible in the capabilities organizations seek. The playbook highlights judgment and taste, business acumen, learning agility and the ability to scope and delegate work to agents. It also describes more T-shaped roles, where deep disciplinary expertise remains important alongside a stronger expectation that people connect their work across the value chain. Engineers, for instance, may become full-stack product builders rather than narrowly focused coding specialists.
When human value moves, jobs need to move, too. Organizations asking people to cross boundaries, create new value and rethink how work happens cannot continue defining them entirely by yesterday’s job descriptions, functions and performance measures. More fluid operating models are needed, with end-to-end value streams, dynamic teams and greater flexibility in traditional structures and incentives.
Building the Next Generation of Human Expertise
This model brings another challenge. If AI handles more of the routine work that traditionally helped people accumulate experience, organizations need to rethink how people build expertise. Judgment will not automatically emerge later in a career. The learning system must continue developing people who can generate new judgment.
“If we don’t invest in early career, we won’t have leaders and experts for the future to continue to guide and govern the way our company works,” George says.
Microsoft explicitly argues for redesigning early-career roles rather than automating them away. It warns that eliminating junior work could hollow out the apprenticeship pipeline that develops future experts. Its PRAISE program matches early-career engineers with experienced mentors in a two-way learning model: senior colleagues teach craft while junior colleagues contribute AI fluency.
That signals an important shift in how organizations approach development. As AI takes on some of the work through which experience once accumulated naturally, companies must deliberately create the experiences that help people build judgment.
A learning system has two responsibilities. It must capture existing organizational knowledge while continuing to develop people who can create knowledge the organization does not yet have.
The Human Operating Model for a Moving Frontier
A frontier firm operates as a human-AI learning system. People contribute judgment, context and experience. AI captures and shares more of that knowledge, making new capabilities possible. People use those capabilities to improve decisions, redefine what good looks like and create the next layer of knowledge the system has yet to learn.
George described an emerging role in which “humans will basically aim the machine, train the machine and own and govern the machine.” She then raised the question Microsoft continues to explore: “What is the human operating model and org structure that will support that?”
This may be the most significant unresolved question in Microsoft’s Frontier Firm Playbook. The loop has a clear logic. Building the structures, roles, development paths and management systems that let it work continuously and at scale is the harder challenge for large organizations.
If organizations succeed, human value will keep evolving beyond a fixed set of tasks or expertise. It will move with the frontier itself, from what the organization already knows toward what its people can discover, develop and teach next.
