New AI Adoption Lessons from Microsoft, Duolingo, Moderna

If there is a single through-line from this year’s WSJ Technology Council conversations, it is this: AI payoffs come from operating models that force alignment, compress decision cycles, and turn “seeing is believing” into a repeatable muscle.

Here are the key case studies you need to know about.

​Microsoft: Kaizen the Enterprise, Not Just the Model​

Speaker: Carolina Dybeck Happe, EVP & COO, Microsoft

Microsoft’s internal AI transformation reads like a case study in disciplined change.

The playbook starts with people, process, and technology working together. Carolina brings leaders into three-day Kaizen workshops to secure cross-functional alignment before a single requirement is coded.

Leaders from sales, finance, operations, security, and product agree on the one problem everyone prioritizes and commit to sticking with it.

The work then goes to the real process.

In the quote-to-cash overhaul for the top 500 customers, a messy 230 steps shrank to 37 value-creating steps. AI agents then automated about 75 percent of those.

The outcome was faster onboarding, stronger security, higher quality, and freed capacity to transform the next process.

Microsoft productized the wins.

The commerce team’s AutoPR agents for high-volume security maintenance were published to an internal agent library, and 25 engineering teams adopted them.

​Change management​ follows a traffic-light strategy: double down on early adopters, convert the cautious majority with proof, and avoid over-investing in active resistors. The mantra is to show results, not just talk about them.

📝 Lesson: Treat AI as an operational system. Alignment, process mapping, agentized automation, and internal reuse make transformation stick.

​Duolingo: AI-First Means Building What Was Previously Impossible​

Speaker: Severin Hacker, Co-founder & CTO, Duolingo

Duolingo’s AI-first memo in April 2024 was a product argument. The company asked: how can AI let us build what could not exist before? The analogy is a restaurant building a mobile app versus Uber being invented because smartphones existed. In Duolingo’s case, that breakthrough is a one-on-one tutor at scale.

The flagship example is Video Call with Lily, an AI conversation partner that finally works because the underlying models do.

Adoption has been strong. On the supply side, Duolingo launched more than 100 courses in a year, a feat that previously would have taken decades. The north star is teaching better and faster, not just adding AI everywhere.

Internally, adoption lagged until leadership solved the real constraint: time.

The company measured usage and instituted “AI Fridays” for exploration, plus targeted mandatory sessions.

Result: near-universal tool adoption across engineering. Speaking of measurement, Duolingo avoids vanity metrics. AI now writes 30% to 40% of code in some places, but coding is only about 20% of engineers' time. The company tracks outcomes such as experiment velocity and shipped impact.

📝 Lesson: AI-first is a product standard. Fund what was impossible before, create protected time to learn, and track real outcomes.

​Moderna, TIAA, Microsoft: Org Design Is the Feature​

Speakers: Tracy Franklin (Moderna), Sastry Durvasula (TIAA), Sophia Velastegui (Microsoft)

Across industries, AI is redrawing organizational models and changing the work inside them.

• ​Moderna merged HR and Technology​ to architect the flow of work. Work planning replaces siloed workforce and systems planning. With democratized AI access and robotics in manufacturing, roles continuously evolve. The goal is speed to patient and faster innovation. This week, Sophie Wade also ​highlighted​ Moderna as a leading case study of integration, noting that other organizations are beginning to follow this model.
• TIAA integrated Technology and Operations. The client experience and the operational engine sit in one organization, enabling end-to-end AI at scale. Fraud prevention for older adults, a two-billion-dollar annual loss problem, now benefits from unified front, middle, and back office teams.
• Microsoft centralized AI under the CTO, partnered tightly with the Chief People Officer, and embedded technology advocates in each business unit. After decades of AI work, the shift from software company to AI company required governance, infrastructure, and change fluency.

Work itself is changing unevenly. 80% of roles are changing by about 20%, while 20% of roles are changing by about 80%. Developers, marketers, and communicators lead the curve.

In regulated contexts, leaders move fast in low-risk zones, keep humans in the loop where stakes are high, and build a secure power grid of policy, privacy, and guardrails that can plug any model into controlled workflows. The cultural shift is to make experimentation the default state.

📝 Lesson: Structure determines speed. Merge where the work meets the customer. Build a shared grid of governance and infrastructure. Train for continuous change.

The Bottom Line

I've said it before, and I'll say it again: all our ​client work​ shows that AI transformation is a cultural and operating model challenge.

The leaders who succeed hard-wire alignment, ship visible wins quickly, and design organizations where product, operations, and people move as one system.

This week, try these five moves:

1. Pick one process everyone agrees is painful and high-impact. Run a three-day workshop with all owners, map the real flow, and isolate the value-creating steps. ​(We can help you with this.)​
2. Assign a builder to the room. Commit to one agent that eliminates at least half of the non-value work.
3. Publish the win. Document the agent and patterns in an internal library and recruit two other teams to reuse it.
4. Protect learning time. Block half a day this Friday for hands-on AI exploration on real work.
5. Measure outcomes. Choose one customer-visible metric, such as time-to-onboard, fraud catches, or experiment velocity, and report back in two weeks.

I track how AI is reshaping organizations, bringing you the news and updates that matter most for scaling AI successfully.  

This week:

AI EXPERIMENTATION TRAP

​HBR: Turning AI Experiments Into Value​

• Failure rate: MIT Media Lab/Project NANDA found 95% of gen AI investments yielded no measurable returns, with most pilots concentrated in sales and marketing rather than core operations.
• Trap repeated: Like digital transformation, leaders are “letting 10,000 flowers bloom” without linking experiments to real customer value, resulting in scattered, under-resourced projects that don’t scale.
• Path forward: Focus AI experiments on serving customers better, use frameworks like intensity–frequency–density (IFD) to prioritize, and build “ninja teams” with senior backing to scale proven pilots.

🚀 Prompt: Ask your team to pick one AI experiment that is clearly improving customer experience or efficiency, and one that feels like busywork. Discuss how to double down on the first and rethink the second.

AI SKILLS GAP

​78% of Tech Roles Now Need AI​

• AI pervasive in ICT: 78% of 50 roles across G7 now require AI skills, with 7 of the 10 fastest-growing jobs AI-related (AI/ML Engineer, AI Risk & Governance Specialist, NLP Engineer).
• Critical gaps: Demand for AI Governance (+150%) and AI Ethics (+125%) shows a rising need for skills at the intersection of tech, law, and ethics. Specialized skills like AI security (+298%) and multi-agent systems (+245%) are surging.
• Human strengths prioritized: While technical skills lag, organizations stress communication, collaboration, and leadership to ensure responsible adoption. Consortium members commit to reskilling 95M people over the next decade.

🚀 Prompt: Ask your team how you can pair AI upskilling with stronger collaboration and leadership behaviors so that technical growth is matched with human strengths that build trust and resilience.

AI IN HR

​How IBM Rebuilt HR with AI​

• IBM centralized 30 HR chatbots into AskHR, enforced adoption by shutting off traditional HR channels, and endured a drop in eNPS from +19 to –35 before rebounding to +74.
• Today, AskHR operates in 52 languages across 30 domains, handling 11.5M transactions yearly, achieving 100% manager adoption and cutting IBM’s HR budget by 40%.
• Key lessons: fix processes before automation, start small on high-volume “moments that matter,” double down on domain expertise, and align culture with tech adoption.

🚀 Prompt: When introducing AI into team workflows, ask how the change impacts employee trust and be ready to adjust based on their feedback.

💨 Quick Read:

• ​Anthropic Index Shows Global AI Divide:​ In the US, 40% of employees now use AI at work, doubling in two years. Task usage is shifting: coding dominates overall, but education and science grew from 9% to 12% and 6% to 7%. Directive automation now makes up 39% of interactions, overtaking augmentation. Contextual data access is emerging as a key adoption bottleneck.
• ​Workday Bets Big on AI With Sana:​ Workday acquires AI firm Sana, making it the “front door” for HR and finance tasks while also launching Workday Data Cloud, illuminating agents, and building a developer platform.
• ​Fed Cuts Rates, Signals More Ahead:​ Fed reduced rates by 0.25%, the first cut since last year, citing weak job growth of only 22,000 jobs last month. Projections: two more cuts in 2025, one in 2026, reflecting a shift in focus from inflation to labor market weakness.