AI Collaboration Is the Work: Extending Jaime Teevan’s Microsoft 365 Community Closing Keynote
Jaime Teevan describes AI and the future of work as a socio-technical evolution, noting that “the technology and the organizational structures that we need to make this work are co-evolving, and we’re right in the middle of figuring all of that out.” Technology and organizational structure are indeed evolving together, often unevenly, occasionally in conflict, and rarely with clear intent. That said, the current moment demands a more forceful interpretation. AI is not simply augmenting work. It is reshaping how intent becomes action, how context is negotiated, and how authority is exercised. The friction many organizations feel is not confusion about tools. It is the exposure of long-standing structural weaknesses that previous generations of technology masked rather than resolved.
The temptation remains to measure progress through personal productivity. Teevan points to increases in document creation and reductions in email time as early indicators, noting that with Copilot, people are creating “10% more documents” and spending “11% less time in email.” Those metrics are easy to capture and easy to celebrate. They are also misleading. Productivity gains at the individual level introduce coordination costs that rarely appear in dashboards.
When more content is produced, more content must be interpreted, validated, reconciled, and acted upon. The system absorbs the excess. In practice, this creates decision-making latency and erodes shared understanding. I have been describing this pattern as the AI tax: local efficiency purchased at the expense of systemic coherence. The issue is not whether AI makes individuals faster. It is whether the organization can metabolize the output without increasing friction.
Teevan herself acknowledges the limits of this approach, arguing that “most of the way that people are using AI right now is for personal productivity,” and calling it “the most boring way possible” to use the technology. She goes further, warning that “naive use of AI for personal productivity can actually harm group productivity.” That observation should be treated as a leading indicator, not a footnote. When organizations optimize locally, they externalize coordination costs. Those costs do not disappear. They accumulate in the system.
This is where Teevan’s emphasis on collaboration begins to intersect with a deeper structural reality. She reminds us that “work is a team sport,” a point that often gets lost in tool-centric discussions. I would take it further. Work is a continuous negotiation—of goals, of constraints, of meaning. Her framing of prompting, grounding, and judgment as the core primitives of interacting with AI aligns with this view, but these should be understood less as technical actions and more as organizational behaviors.
Prompting becomes the act of aligning intent across participants, human and machine. Grounding becomes the arbitration of context, where competing versions of reality are surfaced and reconciled. Judgment becomes the locus of authority, where decisions are made, and ownership is assigned. Teevan notes that these primitives are “inherently collaborative concepts,” tied to alignment, information asymmetry, and accountability. AI forces these dynamics into the open, removing the ambiguity that organizations have historically relied on to maintain momentum.
The shift Teevan describes from individual productivity to group collaboration is real, but it understates the magnitude of change. The more consequential transition is from static artifacts to dynamic flows. For decades, organizations relied on documents to stabilize knowledge. Reports, presentations, and formal outputs acted as checkpoints in the work process. They created a sense of completion and a shared reference point for action. AI disrupts this model by accelerating the creation and iteration of content to the point where the artifact loses its centrality.
Teevan captures this shift when she observes that “knowledge is being created and stored in chat histories, meeting transcripts, and conversations—not just formal documents.” Knowledge now lives in motion—in conversations, in iterative exchanges with models, in ephemeral outputs that inform decisions without ever becoming formalized. This aligns with arguments I have made for moving beyond document-centric knowledge management toward context-centric systems (see Why Your Organization Doesn’t Need a Knowledge Management Vision). The implication is not just technical. It is epistemological. Organizations must redefine what counts as knowledge when it is no longer anchored in stable artifacts.
Teevan’s observation about conversational knowledge points in the right direction, but the operational consequences require more attention. Systems designed to manage documents are poorly suited to managing flows of context. Traditional repositories, taxonomies, and contribution models assume that knowledge can be curated after the fact. In an AI-mediated environment, knowledge must be captured, indexed, and made retrievable in near real time. The ability to reconstruct context—who contributed what, under which assumptions, and with what implications—becomes a differentiator. Without that capability, organizations will generate insight faster than they can retain it. This is where knowledge management re-emerges, not as a support function, but as a core operational discipline embedded in workflows.
The limitations Teevan identifies around socially intelligent AI expose another layer of the challenge. She notes that current systems “don’t understand turn-taking,” “don’t understand social roles,” and fail to account for organizational structure. This is not a minor limitation. It is foundational. Current models operate on patterns. Organizations operate on trust, power, and accountability.
Bridging that gap requires more than better interfaces. It requires a layer that manages identity, authority, and interaction across human and machine actors. In my own analysis of agentic systems, I have described this as the emergence of an agentic control plane—an orchestration layer that governs how agents act, what they are allowed to do, and how their actions are audited. Without such a layer, AI remains a tool embedded in existing structures. With it, AI becomes an active participant in work, requiring governance models that most organizations have yet to define.
The erosion of boundaries that Teevan outlines: spatial, temporal, linguistic, and scale, further complicates this transition. She highlights how AI can overcome “spatial boundaries,” “temporal boundaries,” and even “language and knowledge boundaries,” enabling new forms of collaboration. That erosion is already reshaping work. Remote work removed the requirement for co-location. AI removes the requirement for synchronization. Work can now happen asynchronously, across time zones, languages, and organizational units, with AI acting as a mediator.
What disappears in this process are the implicit coordination mechanisms that once held work together. Meetings, schedules, and physical proximity created alignment through shared experience. In their absence, coordination must be designed explicitly. This is where many organizations falter. They deploy AI tools without redesigning the coordination structures they disrupt. The result is not increased efficiency. It is increased entropy, where more activity produces less clarity.
Teevan’s call to reimagine processes as AI-native speaks directly to this issue. She frames this as one of three pillars, alongside socially intelligent tools and reimagined knowledge capture. The idea is straightforward. The execution is not. Most organizations are layering AI onto existing workflows, automating steps without questioning the process structure itself. This produces faster versions of outdated work. AI-native processes start from different premises: work is asynchronous, knowledge is continuously evolving, decisions are iterative, and agents can act. Designing processes around these premises requires abandoning assumptions that have guided organizational design for decades. It requires moving from linear workflows to adaptive systems that can respond to changing context in real time.
The failure to make this shift is already visible in enterprise AI deployments. Initial enthusiasm gives way to plateauing returns as coordination costs rise and the limitations of existing processes become apparent. This pattern aligns with broader observations from my State of AI work, where the transition from experimentation to operationalization reveals gaps in governance, integration, and organizational readiness. The technology advances rapidly. The organization adapts slowly. The tension between the two defines the pace of progress.
Underlying all of this is a fundamental reframing of what the future of work demands. Teevan states clearly that “human collaboration has been a foundational part of work from the very beginning, and that is not changing.” That point deserves emphasis. The dominant narrative around AI emphasizes automation—doing more with less, replacing human effort with machine capability. That narrative misses the point.
The primary challenge is not automation. It is alignment. As work becomes more distributed, more asynchronous, and more mediated by AI, the ability to maintain shared understanding becomes the critical constraint. AI can assist in generating options, synthesizing information, and accelerating execution. It cannot resolve ambiguity or establish trust. Those remain human responsibilities, mediated through organizational design.
Teevan’s argument that collaboration remains central is correct, but it should be extended. Collaboration is no longer a byproduct of work. It is the work. The systems organizations build must support continuous alignment across participants who may not share time, space, or even language. This requires new metrics. Measuring individual productivity is insufficient. Organizations must assess how effectively they align intent, integrate context, and execute decisions across the system.
The path forward is not ambiguous, even if the details remain unsettled. Organizations need to make explicit choices about how authority is defined in environments where agents participate in decision-making. They need to invest in context management as a core capability, ensuring that knowledge can be captured and reused effectively. They need to redesign processes from first principles, rather than layering AI onto legacy workflows. And they need to develop metrics that reflect system-level outcomes rather than individual output.
Teevan provides a clear articulation of the direction. The challenge now is operational. Organizations that treat AI as a tool for individual productivity will see limited gains and growing friction. Those who recognize AI as a catalyst for rethinking how work is coordinated, how knowledge is managed, and how decisions are made will move beyond incremental improvement. They will redefine how work happens.
That distinction will shape the next phase of the future of work.
For more serious insights on AI, click here.
For more serious insights on strategy and scenarios, click here.
Did you find AI Collaboration Is the Work: Extending Jaime Teevan’s Microsoft 365 Community Closing Keynote useful? If so, please like, share or comment. Thank you!
All images generated via AI from prompts by the author, inspired by photos from the event.
Leave a Reply