Harvard's Raffaella Sadun on why it's so hard to become an AI-first organization

Sadun & Wood (WorkLab) — why it’s so hard to become an AI-first organization

▶ Watch on YouTube · Microsoft · 32:44

The journey to becoming a Frontier Firm requires leaders to reimagine every aspect of their organization - quickly and without a playbook. Harvard Business School’s Raffaella Sadun joins host Molly Wood to explore what makes a Frontier Firm, why reskilling is so hard, and how leaders can build cultures that thrive on experimentation and human-AI collaboration. Discover actionable strategies for navigating uncertainty, driving transformation, and empowering every employee to become an AI-enabled inventor.

TL;DR

A 33-minute episode of WorkLab, Microsoft’s podcast, published 24 November 2025 and recorded on location at Harvard Business School. Host Molly Wood interviews Raffaella Sadun, Charles E. Wilson Professor of Business Administration at HBS, co-chair of the Project on Managing the Future of Work and co-lead of the Digital Reskilling Lab, and a research leader in Microsoft and HBS’s joint Frontier Firm initiative. The episode’s thirteen chapters reduce to five claims:

1. A Frontier Firm is one that invests in learning. “A firm that first and foremost understands that this is the time to learn… a firm that is investing in the process for strategic experimentation and for taking the learnings from this experimentation back into the organization.”
2. There is no playbook, but there may be a playbook for learning. “The scary part is there is no playbook. I think we just have to accept that.” What may emerge instead is the scientific method applied to adoption: hypotheses, measurement, and “distinguishing between fluff and reality.”
3. Every firm is on its own J-curve. The present feels like the Solow paradox of the computer era (“computers everywhere except for the productivity statistics”), but “here it feels like every firm is in its own J curve.” She expects massive divergence: “the same occupation at firm X will be very different from the same occupation at firm Y.”
4. Reskilling is “bloody hard” because “we are asking our workforce… to change who they are.” She recaps the five paradigms from her HBR article Reskilling in the Age of AI, with middle managers as the hinge.
5. Build the adoption stage into the experiment. Experiments run “so brilliantly, but so divided from the rest of the organization” met “almost a dejection, as if it was a foreign object” when it came to scaling.

Key claims

What is different this time

Sadun has studied management since the early 2000s. The last three years look like past technology waves: excitement, then frustration that productivity has not moved. What is new is the promise: AI can summarise and give people access to “tacit knowledge that is typically exclusive to experts,” making knowledge “available and usable and potentially productive to everyone.” The difficulty is how to bring that knowledge into what organisations actually do, and the answer is firm-specific: “you really need to think about your own firm’s specific way of incorporating this new technology into what you do, how you create value, how your workers are organized, how people think.”

Three stages, with strategy first

Her sequence for a successful reinvention:

1. Clarity on what the firm is for. “How are you going to be differentiated? What is the value that you create?” This sets the strategic boundaries inside which experiments run — better product, better service, or lower cost — so that ideas “bubbling up everywhere” can be made sense of. Later she adds a limit: “At some point there can be also too much experimentation,” so experiment “within one important axis of your strategy, or maybe 2 or 3.”
2. Buy-in and adoption. Every major technology shift makes some parts of the organisation feel like winners and others fear being replaced, so the second stage is bringing the organisation along, so that new ideas “don’t just stay with the lab in the firm or with a few people who really see the future.”
3. Compliance and metrics — “but we are very far from that step now.”

Leadership style should follow the phase. Early on, “a leader needs to be close to the action,” because value creation is being redesigned and being near it gives credibility. In the diffusion phase the leader’s job is to find who will build buy-in: skeptics and enthusiasts are “not necessarily correlated with their technical expertise. It’s more of a mindset.” And a warning on measurement: “it can potentially backfire to create metrics that are meant to capture compliance, when you are at a stage in which you are creating new processes.” Early metrics should sustain exploration.

Who runs the experiments, and how they reach scale

The experimentation team needs people who understand “how the technology would get used in the organization or by the consumer,” not only the technology, because those close to the technology “tend not to see all the frictions that might prevent that beautiful invention from actually being used by people.” The capability that matters is iterating quickly on what works. Then: “you have to build that third stage of adoption already in the early stages.” She calls it game theory: start from the end state of mass adoption and ask what inputs, learning and checkpoints are needed to get there, “reverse engineering not just the technology, but how people and organization evolve.”

Her three steps for a CEO who wants to start: a strategy session (“first thing, not months”) to name three things to do; an experimentation team chosen to include “the technical, the human, the potential resistance”; and communication of what is being tried and what it means for the organisation. For an employee: recognise you are “an inventor,” get some technical skills, check that your work fits the organisation’s direction, and join a coalition, because enthusiasts “doing their own thing” produce ideas that “never get compared or aggregated.”

The agent as an eager apprentice

On Microsoft’s framing that every employee becomes a manager of agents: “Even today, it’s not like everybody is a good manager.” An agent is “potentially incredibly knowing and has infinite memory,” but “doesn’t always get it right… has a hard time telling you when it doesn’t know things, and needs specificity.” So it is less an employee than “a very eager apprentice,” and the manager’s skills are asking the right questions, knowing what can and cannot be delegated, and making sense of the answers. She agrees with Wood that the same skills would improve the management of people.

Reskilling: five paradigms

From the HBR article she co-wrote, recapped in order:

1. A strategic imperative. Start with the business rationale, which also gives employees some certainty that there is a future for them.
2. Not just an HR problem but a shared responsibility of the whole organisation, especially the layer reporting to the CEO.
3. The middle-management layer. It holds “incredible coaches and motivators” and others who see reskilling as “an anathema… the additional thing I have to put on my plate.” Her lever is incentives: make mentoring and coaching part of middle managers’ role and evaluation, so that “the creation of human capital in your firm… is something that is seen as currency in the organization.”
4. The employee’s view. Reskilling is a cost to the employee; “right now, people only see the cost.” What is the promise in career, incentives and direction, inside the firm or outside it?
5. It takes a village. Pool knowledge across supply chains and ecosystems to reskill at scale.

On method: programmes that combine hard and soft skills, run in cohorts, and are part of a journey rather than an episode work better than self-paced platforms. Training platforms that “look like Netflix” assume people are motivated to learn on their own: “That’s not true. We are adults, and adults are not necessarily interested in learning new things or admitting that they need to learn new things.” Firms already spend (Wood cites BCG’s 1.5% of total budget); the question is whether they spend it well, again tested “with a scientific mindset… It’s not just enough to do a pilot. You want to have hypothesis. You want to have measurement and you want to learn.”

Divergence, and hiring for learning

Because investments are firm-specific, “we are going to see massive divergence across firms” in how the same job is done, at least in the short and medium run. For employees, the match between person and firm becomes essential, since “three years from now, your job will be different. This is not for everyone.” For hiring: “you are not hiring for the skills that people have today, but you’re hiring people for their ability to learn new skills,” which businesses “often conflate with experience.”

Dynamic-capabilities reading

• digital-seizing/rapid-prototyping — the Frontier Firm is defined by its investment in strategic experimentation: a deliberately composed experimentation team, hypotheses and measurement, fast iteration, and experiments bounded to one to three axes of the strategy.
• digital-transforming/improving-digital-maturity — reskilling the existing workforce is the second half of the episode: the five paradigms, cohort-based programmes mixing hard and soft skills, and hiring for the ability to learn rather than for current skills.
• contextual/internal-barriers — the barriers are named: adults who will not admit they need to learn, middle managers who treat reskilling as extra load, compliance metrics introduced too early, and lab experiments rejected as “a foreign object” at scale.
• strategic-renewal/organizational-culture — “no playbook” means the approach must fit “our culture and our people”; making coaching a rewarded part of middle managers’ role and treating human-capital creation as “currency” are culture changes.

How it connects

• enterprise-ai-adoption — the three stages (strategy, buy-in, compliance), leadership by phase, building adoption into the experiment, and divergence across firms. The same show’s Miller episode gives a structural answer (a frontier unit) to the same question.
• micro-productivity-trap — the Solow paradox restated, and the J-curve made firm-specific; compare Brynjolfsson. Dutt et al. name the trap that a strategy-first approach is meant to avoid.
• durable-skills — hiring for the ability to learn; adults’ reluctance to admit they must learn as the obstacle to reskilling.
• agent-oversight-and-delegation — the agent as an eager apprentice that cannot say when it does not know, and the manager’s job of deciding what to delegate.
• Krishna (IBM) — a CEO’s case against unbounded experimentation, alongside Sadun’s strategic boundaries.
• Dell’Acqua et al., The Cybernetic Teammate — a field experiment Sadun co-authored, on AI in teamwork at Procter & Gamble.

What was actually ingested

The full episode (32:44, length_seconds: 1964), from a human-curated caption track; one ASR-type error was fixed at acquire (“solo paradox” → Solow paradox). The closing show credits are left out of the summary. The HBR article discussed is identified at acquire as Tamayo, Doumi, Goel, Kovács-Ondrejkovic & Sadun, Reskilling in the Age of AI (HBR, 2023); it is not itself in the wiki. No stills: a two-person studio conversation.

Linked entities and concepts

• Raffaella Sadun — guest; entity created in this ingest (second source after the Cybernetic Teammate paper).

• Microsoft — publisher (WorkLab) and partner in the Frontier Firm initiative with HBS. Updated in this ingest.

• enterprise-ai-adoption, micro-productivity-trap, durable-skills, agent-oversight-and-delegation — updated in this ingest.

• dynamic-capabilities — the four cells tagged above.

• Boston Consulting Group — source of the 1.5%-of-budget training figure the host cites.

Dangling (single-source mention, deferred): Molly Wood (host of several WorkLab episodes, but author: is the channel), Harvard Business School.

Scope and reliability

An interview, not a study. Sadun reports patterns from her research and fieldwork but cites no new data in the episode; the one number (1.5% of budgets on training) is BCG’s, cited by the host. The publisher is a partner in the research programme the episode promotes, and “Frontier Firm” is Microsoft’s term. The episode is from November 2025, eleven months before this ingest, and was ingested in a batch with October 2026 sources. Its value is a senior management scholar’s account of sequence (strategy, then buy-in, then compliance) and of why reskilling fails, both consistent with her published work.