Stanford University's SALT Lab analyzed 249,834 real Claude.ai conversations to answer three questions: what work people bring to AI, what control they retain over the process, and where the collaboration breaks down. The findings challenge some of the most common assumptions executives hold about AI adoption.

The most striking finding: 56% of conversations involving actionable tasks were consequential or hard-to-reverse work. Legal guidance. Financial decisions. Strategic choices. People are not using AI for trivial tasks — they are bringing their most difficult problems to it, often without any formal organizational sanction or governance framework in place. Legal and financial guidance had the highest Task Criticality Index at 2.14.

And yet — the human remains in control. 72% of conversations were human-led, with AI assisting rather than directing. Users typically adapted AI outputs rather than using them verbatim. Higher-stakes tasks nearly doubled conversation length and engagement. And friction — which appeared in roughly half of all conversations — was often productive: users who encountered resistance from the AI emerged more engaged and more capable.

The central tension the study surfaces: does working with AI build human capability, or substitute for human effort? Active teaching — where users explain context, correct outputs, and shape results — occurred in 67% of conversations. But the benefits were unevenly realized. The executives who understand this dynamic will design better governance. The ones who don't will discover the gap the hard way.

Key Takeaways

• People are already using AI for consequential, high-stakes work — whether organizations have sanctioned it or not. The governance gap is real and it is already being filled informally.
• 72% of conversations are human-led. The narrative that AI is taking over is not supported by behavioral data at scale. Humans are still in the room — the question is whether the room has the right rules.
• Friction is not failure. Productive friction — where AI challenges or resists — increases human engagement and capability. Optimizing for seamless AI interactions may be removing something valuable.
• AI benefits are unevenly distributed. Employees who already know how to use AI effectively get more capable. Those who don't risk falling further behind. AI literacy is a strategic investment, not a nice-to-have.
• The governance imperative: the highest-risk work is being brought to AI first. Organizations that are still designing their governance framework are already behind the behavior curve of their own employees.

Source: Shao, Y., Zhao, D., Padmakumar, V., Wang, J., & Yang, D. (2026). Human–AI Collaboration at Scale: Task Criticality, Agency, and Friction Across 250,000 Conversations. Stanford University SALT Lab. https://arxiv.org/abs/2608.human-ai-collaboration-at-scale

Hosted on Acast. See acast.com/privacy for more information.

In this second and final episode of the Management Illusions series, Kenza and Pascal cover the remaining three illusion domains — and the findings are the ones most executives will recognize from their own boardrooms.

Illusion Domain 3 covers the ROI stage. The downsizing illusion — the most prevalent and most damaging of all ten illusions according to the expert survey — is the belief that AI agents will enable significant workforce reductions. Ford, Klarna, and Salesforce all had to rehire thousands of workers whose AI replacements showed unreliable performance. Up to 30% of workers laid off due to AI may need to be rehired within three years. The productivity illusion reflects the well-known productivity paradox: measured productivity improvements from agentic AI trail perceived and expected improvements. Accelerating one process creates little firm-level value if downstream processes become bottlenecks.

Illusion Domain 4 concerns how firms organize for AI. The bottom-up illusion is the belief that AI champions or ambassadors at the mid-level can drive the transformation. No firm has succeeded with this model alone: champions lack time, technical authority, and formal power to change how work is organized. The agent=employee illusion is the urge to anthropomorphize AI agents — giving them human names, placing them on org charts — which early evidence suggests may undermine managerial oversight and accountability.

Illusion Domain 5 addresses strategy. The master plan illusion is the belief that firms need a firm-wide agentic AI vision and roadmap. Agentic AI is an instrument for executing a business strategy — not a strategy in itself. The preparation illusion comes in two opposite failure modes: starting without sufficient foundations, or waiting indefinitely for perfect conditions that never arrive.

Key Takeaways

• The downsizing illusion is the most damaging of all ten illusions. Up to 30% of workers laid off due to AI may need to be rehired within three years. Deploying AI agents requires substantial human resources to build, test, maintain, and oversee them.
• Productivity gains in individual tasks do not automatically translate to firm-level productivity. Accelerating one process is worthless if downstream processes become bottlenecks.
• The honest reframe: agentic AI is a long-term investment in competitiveness — not a short-term cost reduction lever.
• The bottom-up illusion: AI champions without dedicated time, technical depth, and formal authority produce monthly update meetings, not transformation. Successful firms combine top-down enablement with bottom-up initiative.
• In one firm of 30,000 employees, successful transformation required approximately 60 dedicated full-time employees centrally — with HR as a close partner, because scaling AI means reorganizing work.
• A firm-wide agentic AI master plan puts the cart before the horse. What firms need is a governance and infrastructure framework within which functions can deploy AI where it genuinely helps.
• The final test for any AI investment: if you removed the word 'AI' from the proposal — would it still be a compelling business case?

Hosted on Acast. See acast.com/privacy for more information.

In this first of two episodes based on a major new study by Johannes Habel and colleagues — validated through a survey of 167 technical experts — Kenza and Pascal unpack the four management illusions that undermine agentic AI at the use case development and deployment stages.

Illusion Domain 1 covers two illusions around developing use cases. The agent-first illusion is the belief that AI agents can and should eventually enter virtually any business process. In reality, only about 10% of automations in a large-scale transformation actually require LLM-based agents — the rest can be handled by simpler, more deterministic tools. The end-to-end illusion is the expectation that a single AI agent can reliably navigate complex multi-step processes. Today's agents require highly specialized scope and human orchestration; broader process improvement emerges by combining multiple agents with other technologies and humans in the loop.

Illusion Domain 2 covers the deployment stage. The testing illusion is managers' systematic underestimation of the effort required to evaluate and safeguard agent reliability. AI agents that perform well in demos regularly fail in unattended production — often silently — and the availability of domain experts, not technical experts, is usually the bottleneck. The scaling illusion is the assumption that a successful pilot can be replicated quickly and cost-effectively. In practice, processes that look similar often differ in inputs, IT systems, and data quality — and the human resource challenges of scaling AI agents are as significant as the technical ones.

Key Takeaways

• AI agents should be the last resort for automation, not the first. Only about 10% of identified automations in one large-scale transformation required LLM-based agents.
• The end-to-end illusion: today's AI agents cannot reliably navigate complex multi-step processes. The right approach is specialized agents for clearly defined subtasks, orchestrated alongside humans and other tools.
• AI agents that pass demos regularly fail in production — often silently. A voice agent in automotive called former customers to reactivate them; it performed significantly worse than human agents and caused many opt-outs.
• The testing burden falls on domain experts, not technical experts. Their availability — not the technology — is usually what limits deployment timelines.
• Scaling is not replication. A pilot designed to prove feasibility is rarely designed to scale. The second implementation almost always takes longer and costs more than expected.
• Scaling AI agents ultimately means scaling human agents too — reorganizing roles, building new skills, and securing employee buy-in.

Hosted on Acast. See acast.com/privacy for more information.

Most executive decisions are forecasts in disguise. 'Will this market grow?' 'Will this AI initiative deliver ROI?' 'Will this acquisition integrate well?' These are all probability estimates — and most leadership teams make them without any structured method, relying instead on experience, intuition, and the confidence of whoever speaks loudest in the room.

In this episode, Kenza and Pascal explore two landmark studies on AI-augmented forecasting and what they mean for executive decision-making. The first, by Halawi et al. (NeurIPS 2024), shows that a retrieval-augmented AI system can approach the forecasting accuracy of competitive human forecasters — and that combining AI and human judgment always outperforms either alone. The second, by Schoenegger et al. (2025, co-authored with Philip Tetlock, the world's leading expert on forecasting), shows that AI assistance improves human forecasting accuracy by 24–28% — even when the AI assistant is deliberately biased and overconfident.

The key insight: the benefit is not just informational. It is cognitive. Engaging with an AI perspective — even an imperfect one — forces assumption-surfacing, reduces overconfidence, and makes human judgment measurably more accurate. The Brier Score — a rigorous calibration metric that measures how well-calibrated probability estimates are, not just whether someone was right or wrong — confirms the improvement is real and quantifiable.

For boards making high-stakes decisions under uncertainty, this is not a hypothetical. It is a demonstrated, quantifiable capability that most executive teams are not yet using systematically.

Key Takeaways

• Combining AI forecasts with human judgment always produces better outcomes than either alone — confirmed across thousands of forecasting questions.
• AI assistance improves human forecasting accuracy by 24–28%, measured by the Brier Score — the gold standard for calibration.
• The benefit is cognitive, not just informational: AI forces you to surface assumptions you didn't know you were making and reduces overconfidence even in experienced forecasters.
• AI performs best exactly where boards struggle most — in the 30–70% uncertainty range where nobody is sure and overconfidence is highest.
• Most executive teams use AI to summarize and report. The studies suggest they should be using it to challenge assumptions before the decision is made.
• The practical starting point: pick one recurring decision type, introduce probability language, track outcomes, and score calibration over time.

Hosted on Acast. See acast.com/privacy for more information.

Nearly 9 in 10 organizations report using AI regularly. Only 37% say AI has made a positive contribution to their EBIT. And only 6% qualify as genuine AI high performers. The rest? Optimizing. Automating. Saving costs. Declaring victory.

In this episode Kenza and Pascal ask themselves: when does AI actually grow the top line?

And close with a board checklist and a final provocation:

"Do not ask what AI can do for your current business. Ask what business you can build because AI now exists."

Key Takeaways

• Efficiency asks 'how.' Growth asks 'what, for whom and why us?' These are different questions requiring different organizational conditions.
• Capability without mandate produces demos. Mandate without capability produces PowerPoint. Both without a business model hypothesis produce expensive activity.
• A Copilot license is not a strategy.
• The winners will not be the companies that bought the most AI. They will be the ones that used AI to become a different company while they still had the luxury of choice.

Hosted on Acast. See acast.com/privacy for more information.

In this episode, Kenza and Pascal tackle the hardest part of AI transformation: the gap between having insight and acting on it. Pascal brings a provocation that cuts to the heart of how most companies measure AI ROI: an insight is not an innovation. It only becomes innovation when it changes a product, process, customer experience, or business model.

Key Takeaways

• The last mile of AI is not prediction. It is authority — who owns the decision that needs to change.
• An insight is not an innovation. If no decision, process, product or customer outcome changes, you produced information, not value.
• Process intelligence exposes the difference between how a process is described and how it actually behaves. That gap is usually where the real AI opportunity sits.
• A process map can be political. A timestamp is less diplomatic. Use process evidence to redesign the system, not to publicly grade departments.
• You cannot ask people to challenge the status quo while measuring them only on how reliably they preserve it.
• Data without decision rights is just expensive observation.

Hosted on Acast. See acast.com/privacy for more information.

The episode closes with a line worth writing down: a dashboard can be technically correct and still be organizationally untrustworthy. And another: trust should be earned by the system, not requested by the project team.

Key Takeaways

•      Data governance only works when it protects value and speeds decisions — not when it is purely a compliance exercise people route around.

•      The business owns the meaning of data. Technology owns the enablement. Governance connects the two.

•      AI did not repeal garbage-in, garbage-out. It just made the garbage more eloquent. LLMs can mask data quality problems — which makes them more dangerous, not less.

•      Trust in data must be built progressively: low-stakes decisions first, visible provenance, measurable error rates, and explicit human override.

•      The opposite risk is equally dangerous: executives who trust the data too much. Confidence is not evidence. Mature organizations institutionalize questioning.

Hosted on Acast. See acast.com/privacy for more information.

In this episode, Kenza and Pascal move from technology foundations into the data layer and make the case that most data problems are not technical problems at all. They are cultural, political, and leadership problems. And often, they are the result of organizations not knowing what they actually want to know.

Key topics:

• the DIKW pyramid (Data, Information, Knowledge, Wisdom) as a framework for understanding where companies get stuck,
• why data silos are political artifacts rather than technical accidents,
• how to move from raw data to decisions that actually change something.

Data silos are rarely technical accidents. They are political artifacts,  fiefdoms inside organizations that protect data as a source of power.

Hosted on Acast. See acast.com/privacy for more information.

Key topics:

- architecture choices that enable or constrain AI at scale,

- why disconnected systems create AI blind spots,

- the infrastructure investment needed for production-grade AI

- why cybersecurity is now an expanded executive responsibility.

Pascal's provocation: Legacy architecture is frozen decision history. The question for every board is which decisions are you still living with, and are they still the right ones?

Hosted on Acast. See acast.com/privacy for more information.

Key topics:

• why IT perception inside the organization determines AI success,
• process clarity as a precondition for automation,
• how to make the investment case for technology readiness when boards are focused on short-term efficiency.

AI does not repair broken processes. It industrializes them.

Hosted on Acast. See acast.com/privacy for more information.

In this episode, Kenza continues her conversation with Dr. Abir Haddad, who works extensively in the region, for a ground-level view of what is actually happening. They explore what Europe can learn from the Gulf's approach to AI and why the region is uniquely attractive for startup founders and growth-stage CEOs: direct access to governments as first customers, national AI strategies that actively seek technology partners, and an innovation culture that moves at a speed most European ecosystems simply cannot match.

KEY TAKEAWAYS

• The Gulf is investing heavily in AI ecosystems and talent.
• Start-ups gain easier access to funding and government projects.
• Regulation relies more on guidelines, sandboxes and experimentation.
• Central AI governance coordinates activities across institutions.
• Human oversight and continuous upskilling remain essential. 

Hosted on Acast. See acast.com/privacy for more information.

KEY TAKEAWAYS

• AI regulation is becoming global.
• EU, US and China follow different models.
• The use case determines the actual risk.
• Clear governance enables faster AI adoption.
• Executives must know how AI is governed, documented and controlled.

Hosted on Acast. See acast.com/privacy for more information.

In this episode, Kenza and Naureen explore the full spectrum of AI risk beyond legal exposure: bias, unreliable outputs, data privacy, reputational consequences, and the risk that receives the least attention — gradual, unnoticed reliance. They introduce a practical framework for building risk awareness across the organization, and discuss why silence from employees is often the biggest risk signal of all.

This is also the final episode of Season 2. Kenza thanks Naureen for bringing her perspective to the podcast, and previews what is coming in Season 3: what it actually takes to scale AI inside organizations, and how AI capabilities eventually lead to entirely new business models.

KEY TAKEAWAYS

•      AI rarely creates risk through dramatic failures. It creates risk through gradual trust — outputs that quietly shape decisions before anyone formally validated the model.

•      The risk leaders most often underestimate: AI becoming embedded in decision-making without clear oversight or accountability.

•      A practical framework for managing AI trust: Visible (everyone knows where AI is used), Questioned (outputs are open to challenge), Verified (systems are regularly reviewed).

•      Responsible AI is not just about frameworks. It is about how people act within them.

Hosted on Acast. See acast.com/privacy for more information.

Kenza and Naureen explore why traditional compliance models struggle with AI, what compliance by design looks like in practice, and how the EU AI Act changes what organizations need to prepare for. They challenge the most common misunderstanding: that regulation is designed to stop AI. It is designed to create trust in AI adoption — and companies that prepare early usually find that good compliance is simply good governance.

KEY TAKEAWAYS

•      Compliance works best when legal and compliance teams participate in shaping AI initiatives — not when they review them at the end.

•      The EU AI Act uses a risk-based approach: not all AI systems carry the same obligations. Classification is the first step.

•      High-risk AI systems require documented governance, human oversight, and clear accountability — not just a privacy policy.

•      Organizations that treat the AI Act as a trust-building exercise rather than a penalty-avoidance exercise gain a competitive advantage.

Hosted on Acast. See acast.com/privacy for more information.

In this episode, Kenza is joined by Naureen, a governance and compliance expert with over twenty years of experience in complex regulated environments. Together they explore what AI governance actually looks like when it is designed to enable innovation rather than block it — and why organizations that introduce governance early consistently move faster later, because teams already know the boundaries.

They also discuss the CEO's personal role in making governance work, and why conflicting signals from leadership are often the biggest compliance risk of all.

KEY TAKEAWAYS

•      Governance creates confidence to move. Without it, people slow down because nobody is sure where the boundaries are.

•      The right moment to introduce AI governance is when AI starts influencing real business decisions — not when regulation forces you to.

•      Introduce governance one step before you scale AI, not one step after.

•      CEOs who treat compliance as part of how the company operates — not as a side function — are the ones whose organizations actually adopt it.

Hosted on Acast. See acast.com/privacy for more information.

Once organizations move beyond early AI experiments, a new question appears: Do we have the right people to scale this? Eventually internal learning reaches its limits. You need deeper expertise. But hiring AI talent is one of the most competitive challenges companies face today — and the real question is not just who to hire, but how to build the right combination of people.

In this episode, Kenza and Amanda discuss when external AI expertise is actually needed, why the search for a unicorn AI hire so often fails, and what a healthy AI team really looks like. They explore how to integrate new talent without creating an elite separate group — and how to retain people who have endless options elsewhere.

This is also Amanda's final episode as co-host. She hands over to the next season with a clear message: AI transformation starts with the people who already understand your business.

KEY TAKEAWAYS

•      Hiring too early is as problematic as hiring too late. The trigger should be scaling from pilots to production — not ambition.

•      AI success does not come from one brilliant hire. It comes from a system of people working together.

•      The strongest teams combine technical specialists, domain experts, and operational leaders — and pair new external talent with long-tenured internal knowledge.

•      AI talent stays where they can see their work matter. Meaningful problems and speed of execution matter more than compensation alone.

Hosted on Acast. See acast.com/privacy for more information.

When organizations talk about AI transformation, the instinct is often to hire. But that is usually the wrong starting point. The real question is how to enable the people already inside the business.

In this episode, Kenza and Amanda explore what AI literacy actually means inside organizations and why treating it as a training program rather than a business capability is where most companies go wrong. They discuss the three levels of AI literacy (leadership, operational, and technical), why upskilling existing employees should come before external recruitment, and how co-creation across functions accelerates adoption in ways that top-down rollouts rarely do.

KEY TAKEAWAYS

•      AI adoption is not primarily a hiring challenge. It is an activation challenge.

•      AI literacy only sticks when people apply it to their own decisions and processes - not when it stays in a training room.

•      Leaders who visibly participate in AI learning send a signal the rest of the organization cannot ignore.

•      Early adopters who pull others along are more valuable than any single AI hire.

Hosted on Acast. See acast.com/privacy for more information.

When executives ask about AI transformation, the question is often: what is the right organisational design? Central team or decentralised? The honest answer is: you need both - and the balance shifts as your transformation matures.

In this episode, Amanda and Kenza explore what it actually takes to structure an organisation for AI. They discuss why a dedicated central team is critical in the early stages, how to handle the tension between day jobs and transformation work, and why measuring return on investment in AI is far harder - and more nuanced - than most boards expect.

KEY TAKEAWAYS

• Start with a central team close to the CEO to orchestrate initiatives and set the guardrails. Its importance decreases as the organisation matures.
• People cannot drive transformation on top of their day jobs. Organisations must explicitly free up capacity or accept that performance targets will temporarily take a back seat.
• AI handles structure well. Humans handle chaos. The smartest deployments automate the repetitive 80% and invest in people for the complex 20%.
• Before you automate anything, optimise the process first. Automating a broken process just makes the broken parts faster.

Hosted on Acast. See acast.com/privacy for more information.

In this episode, Amanda and Kenza explore why culture is the single most important lever in any AI transformation and how to move it deliberately. They discuss how a clear strategic vision needs to reach every level of the organization, why middle management deserves far more attention than it typically receives, and how to build a genuine learning culture where experimentation is encouraged and mistakes are not punished.

KEY TAKEAWAYS

• Culture is not the values on the lobby wall. It is how people behave when no one is watching.
• AI transformation cannot be delegated to the CTO alone - every executive owns a piece of it.
• The "frozen middle" is not resistant to change. It is overwhelmed. Organizations must explicitly create space for managers to learn and experiment.
• As long as mistakes carry career risk, adoption will stall. Psychological safety is not a nice-to-have. It is a prerequisite.

Hosted on Acast. See acast.com/privacy for more information.

In this first episode, Kenza and her co-host Amanda introduce the podcast and the question that drives it: why AI transformation has moved from the innovation lab to the executive agenda — and what that shift really means for leaders.

They explore the gap between AI expectation and organizational reality, why most leadership teams are still figuring it out, and what a genuinely useful conversation about AI at the executive level actually looks like. The show is positioned not at the extremes — not deep tech, not distant future — but in the messy middle where real leadership decisions happen.

KEY TAKEAWAYS

•      AI has moved from innovation experiment to strategic priority — and leadership teams are expected to have answers.

•      The real challenge is not technology. It is deciding where to place the first serious bet — and what that means for the rest of the organization.

•      The most valuable AI conversations are the ones that usually stay inside executive rooms. This podcast brings them out.

Hosted on Acast. See acast.com/privacy for more information.