By the Team Flow Institute Research Fellows
December 2025
Executive Summary
Three years into the generative AI revolution, organizations face an unexpected bottleneck: not technological capability, but human readiness. While AI tools have proven their ability to unlock tremendous efficiencies and enable individual flow states through rapid iteration, widespread enterprise adoption remains constrained by emotional resistance, cultural hesitation, and a fundamental challenge with non-deterministic outputs.
Key findings from recent roundtable discussions with our Research Fellows:
The adoption gap is human, not technical – Even organizations with proprietary AI products struggle with internal engagement, despite proven business cases. The core problem is user adoption and willingness to engage, not ROI.
Most applications optimize rather than innovate – The majority of current use cases accelerate existing workflows (work projects compressed from six months to three weeks, research from days to hours), while truly novel applications remain concentrated in scientific domains.
Individual flow unlocks team potential – The use of generative AI technology can enable faster, more independent work. For example, engineers and data scientists experience genuine flow states through “vibe coding” and rapid prototyping. This individual flow state could lead to better team flow.
Non-determinism blocks trust-based deployment – Regulated industries require 100% validation of AI outputs due to unpredictable errors, eliminating efficiency gains and eroding trust in AI tools. Organizations in highly regulated industries such as finance, healthcare, and legal cannot deploy generative AI in high-stakes applications when outputs vary unpredictably. The business case is diluted.
Expertise development is at risk – Accelerating production through AI may inadvertently starve the pipeline of future experts capable of exercising the human judgment that remains essential.
The next frontier is agentic collaboration – Multi-agent systems capable of autonomous research and iteration represent the next phase, though development is proceeding more slowly than anticipated.
Organizations that successfully navigate this transition will recognize that AI adoption is fundamentally a transformation of how people work, requiring new ways of communicating and collaborating, transparency practices, and a reimagining of what skills and roles provide strategic value.
Introduction: Beyond the WordPerfect Stage
Three years after ChatGPT’s public launch, we find ourselves in what Holtz calls “the WordPerfect stage of AI” — using a powerful new technology primarily to do the same old things faster, not yet imagining the true transformative power that can emerge from this powerful new technology.
“We’re essentially where the personal computer was in its early days, which was as a replacement for a typewriter,” explained Team Flow Institute Fellow Shel Holtz. “That’s where AI is right now. All the uses that we see are: can I write faster? Can I find quicker answers to questions? That’s not the end game by any stretch of the imagination.”
This analogy to an earlier technological shift captures the paradox facing organizations in late 2025.
The Real Barrier to Adoption: The Emotional Readiness Gap
“Organizations lack emotional readiness for AI adoption despite technological capability,” Chris Heuer, Team Flow Institute Managing Director, observed, citing a recent panel hosted by a major consulting firm. “Even the advocates of AI transformation face internal resistance with their own AI products. The core problem is user adoption and willingness to engage, not ROI.”
This finding, confidentially confirmed by enterprise AI leaders, contradicts the dominant narrative around AI adoption. While organizations focus on building business cases and measuring productivity gains, the actual bottleneck is human acceptance.
As Heuer noted, “When the layoffs are hitting in the sizes they are, particularly in the technology sector which is the canary in the coal mine for adoption, this is an important bellwether.” Yet the Team Flow Institute Fellows agree that current layoffs stem more from post-pandemic over-hiring corrections than AI displacement. This distinction matters: it means the fear-driven resistance may be unfounded, even as the emotional reality of that fear shapes adoption patterns.
The state of the enterprise adoption gap is well-documented across major research studies. McKinsey’s 2025 State of AI survey reveals that while 78% of organizations now use AI in at least one business function — up from 56% in 2021 — only 39% report enterprise-wide financial impact (McKinsey, 2025). The majority of organizations (75%) lack an enterprise-wide roadmap for generative AI, and less than 40% report that senior leaders understand how the technology can create value for the business (McKinsey, 2024). Deloitte’s 2025 research found that almost three-quarters of companies report their most advanced AI initiatives have met or exceeded ROI targets, with approximately 20% seeing returns over 30% (Deloitte, 2025). Yet this success at the pilot level has not yet translated to enterprise-wide strategic deployment.
As Team Flow Institute Fellow Steve King observed from decades of managing teams: “Humans are complicated. And if you’re running a large group of humans, your life’s incredibly complicated.” This complexity is precisely what organizations hope AI will simplify — but it may also be what makes human judgment irreplaceable.
The Craft Identity Challenge
Resistance to the adoption of generative AI seems to run deepest among those who view their work as craft. “We’re seeing this in creative roles, where love of the craft meets the fear of being replaced by technology.” observed Jen McClure, Senior Research Fellow.
This craft identity challenge appears across domains. Heuer described encountering software developers who resist “vibe coding” (AI-assisted rapid prototyping) because “coding is a craft and they have a craftsman’s mindset.” The enjoyment of writing code line by line — the production work itself — becomes more valued than the meta-cognitive work of architectural design and problem-solving.
The implications extend beyond individual resistance. When those who control critical workflows resist adoption, entire organizations slow down. As Holtz noted from his construction industry experience, “I have a sneaking suspicion that out in the project trailers on construction sites, managers are not encouraging their employees to use these new tools. They’re saying, ‘We have a schedule to meet. Do your job.'”
McKinsey’s research on AI adoption barriers confirms these organizational dynamics, identifying knowledge and training gaps (51%) and regulatory uncertainty (40%) as significant challenges (McKinsey, 2025). The survey found that cultural apprehension and organizational inertia create implicit resistance from business teams and middle management due to fear of disruption, uncertainty around job impact, and lack of familiarity with the technology (McKinsey, 2025).
From Optimization to Novel Synthesis
The Compression of Complex Work
AI’s current sweet spot seems to be synthesizing large volumes of quantitative and qualitative data. The work isn’t fundamentally different from traditional human work, but the speed and comprehensiveness represent a quantum leap. Team Flow Institute Fellow Dr. Michael Wu discussed this in the context of engineering domains: “The members of our teams are a lot more independent than before in their work.” This independence — eliminating some coordination handoffs and frictions between team members — may be as valuable as speed itself. As Wu noted, “An engineer can now do a lot more without dependency on others.”
The Question of True Novelty
Yet Holtz’s challenge remained: “It’s still in pursuit of the same outcomes that took the longer approach before. What can we do that isn’t something that we could do with the old technology?”
Wu offered the strongest counter-evidence: algorithmic breakthroughs that transcend synthesis. “The matrix multiplication algorithm operates fundamentally differently than prior human approaches,” he explained, describing Google’s team at DeepMind using AI to discover methods that eluded the world’s best mathematicians. “People have tried to do 4×4 matrix multiplication with less than 49 operations forever, and they failed every time. So there’s nowhere in the world that has that algorithm. How was AI able to come up with this? It’s actually combining existing knowledge in a novel way, a novel synthesis.”
Other examples of genuine novelty emerged:
- Fusion reactor design: AI-generated chamber designs with novel micro-structures that dramatically reduce corrosion, “defying conventional geometric logic”
- Peer-reviewed AI research: Systems iterating through 20+ paper submissions before publishing, with acceptance at major conferences
- Quantum computing applications: Exploring permutation spaces too vast for human iteration
As Wu summarized: “The language model essentially did some really complex transformation of the problem to a completely different space. You translate an algebra problem into a geometric problem, then translate it to a topology problem, and then use some theorems in topology to solve the problem that didn’t exist in geometry or algebra, and then translate them back.”
This capability—translating problems across mathematical domains to find novel solutions impossible in the original formulation—represents true novelty. The question is whether such breakthroughs will remain confined to scientific domains or eventually transform knowledge work more broadly.
Beyond Human Intuition
While most AI applications optimize existing approaches, some outputs transcend what humans would conceive. Dr. Wu described AI-generated fusion reactor chamber designs that defy conventional engineering logic: “A human would never design the chamber that way. They would design in some geometric way. This almost feels like something that’s grown.”
The result: novel micro-structures that dramatically reduce corrosion, extending reactor operation far beyond what traditional designs allow. The AI wasn’t faster at human-style engineering — it found solutions in a design space humans wouldn’t explore.
These breakthroughs remain concentrated in domains with clear success metrics — physical performance, mathematical proof, experimental validation. Whether such genuine novelty will emerge in less measurable knowledge work remains an open question.
Individual Flow and Team Dynamics
The Emergence of Vibe Coding Flow
The most tangible transformation described by Fellows involved individual flow states enabled by rapid AI-assisted iteration. Wu described his engineering teams: “They’re actually more engrossed in what they’re doing because they’re in this flow state.” The mechanism is “vibe coding” — conversational code generation that maintains creative momentum. “When you have an idea, you can just vibe code it and you generate all the code right away, and you can iterate,” Wu explained. “So you get into this state of flow very quickly and you want it to just continue.”
The Echo Chamber Risk
McClure raised a crucial concern: Can this independence lead to an echo chamber effect? Generative AI and large language models can be very isolating and risk being an echo chamber when individuals work in isolation with AI. Outputs reflect only their input framings and assumptions, potentially amplified. McClure suggested a solution to this: collaborative refinement of AI output with stakeholders before finalizing and basing decisions on the output. This suggests that AI doesn’t eliminate the need for collaborative sense-making — it shifts when and how that collaboration occurs.
The Transparency Paradox
This collaborative approach requires new forms of work transparency, a “Let me show you how I got here. Here is my original query. Here are the additional questions. Here’s how I answered those additional questions,” McClure explained. “I want whoever I’m working with on refining the output to be able to say, ‘Why didn’t you ask this in your prompt?’ or ‘Why did you answer this that way? Generative AI offers a powerful tool to test our assumptions if used in this way with teams.”
This level of transparency — showing not just outputs but the entire prompt engineering process — represents a significant cultural shift. Yet it may be temporary. As Wu noted about calculators and spreadsheets, “In the early days when spreadsheet was created, people literally went and checked the spreadsheet calculation with a calculator. Once you have that trust established, people will say, ‘This is deterministic enough that you don’t need to really show the work anymore.'”
The question becomes: when does AI work transition from requiring full transparency to becoming normalized practice where disclosure is unnecessary?
The Evolution of Disclosure
The Fellows identified an evolution in how AI-assisted work is disclosed and validated. Holtz described using ChatGPT to synthesize 450 bullet points from a SWOT analysis — work that would have taken two days compressed into two hours with validation. “I did not tell the EVP that. I didn’t disclose. I don’t think he would have cared. I think that is kind of an equivalent of ‘I used a calculator instead of crunching the numbers on a piece of paper.’ “
Yet Wu cautioned about shared workflows: “Because these systems are non-deterministic, if you share your work and your colleagues then go and repeat what you do, they may actually come up with a different result.”
The question of when disclosure becomes necessary (or unnecessary) depends on multiple factors:
- The stakes of the work (expense reports vs. medical diagnoses)
- The maturity of the technology (calculator-level determinism vs. LLM variability)
- Organizational culture and risk tolerance
- Whether the goal is learning and improvement or simply production
From Individual to Team Flow
The relationship between individual flow and team dynamics proved more nuanced than simple aggregation. Heuer referenced Stewart Butterfield’s observation that “communications overhead reaches 20% for standard teams, 40-60% for larger teams.” When individuals can complete complex work rapidly through AI assistance, “coordination failures don’t cascade.”
This creates a virtuous cycle: individual flow reduces coordination overhead, which enables more individual flow. Teams become “more engrossed and self-directed when flow states are accessible,” as Wu observed.
Yet McClure’s caveat about isolation risks persists.She suggests presenting AI-generated outputs to stakeholder teams, gathering feedback, and iterating collaboratively — preserving the social fabric of work even while accelerating production.
Team Flow Institute Co-founder Jaime Schwarz framed this as a new collaborative ladder:
1. Solo Flow – individual working alone
2. Augmented Flow – individual + tools
3. AI-Augmented Flow – individual + AI as a “second mind” collaborator
4. Team-Augmented Flow – Team using tools and practices collaboratively to boost collective flow
Heuer noted that “Despite the potential of AI to enhance team collaboration and coordination, some of the early research seems to indicate that AI is failing at the team level, creating more friction than gains. This is why we formed the Team Flow Institute, to ensure that we didn’t introduce the transformative capabilities of AI without first establishing better team dynamics like trust and psychological safety.”
The Determinism Challenge
Why 99% Accuracy Isn’t Enough
The most significant technical barrier to widespread adoption emerged clearly in the Fellows’ discussion: non-deterministic outputs. “LLM outputs are non-deterministic,” Wu explained. “Identical inputs produce different outputs, unlike calculators or spreadsheets. For example, insurance claims processing demonstrates this barrier: 1% indeterminate error rate requires validating all output, eliminating any time savings gained by using AI as a tool.”
This creates an impossible economic problem. If you must check every output to catch the 1% that’s wrong, you’ve eliminated the efficiency gain. As Wu put it: “If I’m going to check everything, I might as well just do it.”
The impact extends beyond efficiency to trust. Organizations in regulated industries — insurance, finance, healthcare, legal — cannot deploy generative AI in high-stakes applications when outputs vary unpredictably. The business case disappears.
This challenge is compounded by the nascent state of AI scaling efforts across enterprises. Most organizations are still navigating the transition from experimentation to scaled deployment — approximately 63% of companies remain in the pilot phase (McKinsey, 2025). In any given business function, no more than 10% of organizations are scaling AI agents (McKinsey, 2025). The gap between experimentation and production deployment stems from several interconnected challenges: data accessibility and quality gaps, cultural apprehension, and the technical complexity of integrating AI with legacy systems.
The Path to Deterministic AI
“Solving indeterminism is a prerequisite for widespread trust-based adoption,” Wu said, pointing to Mira Murati’s post-OpenAI venture. “Mira Murati’s company Thinking Machines specifically focuses on this as a precondition for high-stakes applications. Once resolved, validation requirements decrease and people naturally stop requiring work to be shown.”
The parallel to early spreadsheet adoption is instructive. Users validated Excel calculations with calculators until determinism was established through repeated experience. Only then did showing work become unnecessary.
Wu suggested the timeline: “When we solve the indeterminism problem, that’s when the use of generative AI will be more accepted. That’s when ‘showing our work’ won’t be necessary anymore.” But the technical path to solving non-determinism in language models remains unclear, and may represent a fundamental limitation of current transformer-based architectures rather than a solvable engineering problem.
Human Judgment at the Center
What Cannot Be Offloaded
Even as AI capabilities expand, the Fellows identified coordination functions that must remain distinctly human. Heuer posed the question: “What coordination capabilities should never be offloaded?” The responses clustered around judgment, accountability, and meaning-making:
Accountability: “There’s no way to hold an AI accountable,” Wu observed. “What are you going to do — ban the AI if it makes a mistake? Ultimately it’s the human who worked with that AI or manages the AI who’s going to be accountable.”
Ownership: Schwarz emphasized that “whoever is putting in the input or whoever is generating the output is the one who’s in charge or responsible for those two things, not the LLM.”
Transformation through vulnerability: As Heuer noted from his personal development conversations with Claude, genuine organizational transformation requires human-to-human connection and risk-taking that AI cannot replicate.
Shared mental model formation: Teams must develop mutual understanding together, a process that cannot be fully delegated to AI intermediaries.
Strategic priority reset under uncertainty: “Never offload the capacity to fundamentally question and reframe what you’re coordinating toward when conditions change dramatically,” Heuer emphasized.
The Assembly Line Analogy
Heuer proposed this analogy: “Somebody had to design the assembly line. Somebody had to figure out when this part goes in, that it was needed, and that it served a certain function. That original human thinking and the curiosity of that thinking allows for proper framing.”
The implication: AI automates production work, but human judgment should provide the innovation, the contextualization, framing, and quality evaluation. This returns us to team flow principles: the value lies in human judgment about which problems to solve, how to frame them, and what constitutes a good solution — not in the mechanical work of generating solutions.
The Expertise Development Paradox
Yet a troubling question shadows these conversations: how do people develop expertise if AI handles the foundational work?
King illustrated the risk with a recent client engagement. Junior analysts presented a “payout ratio” of 74% for S&P 500 companies – figure King immediately recognized as impossible under standard definitions. When challenged, they discovered their analysts had redefined the term without disclosure. “These were smart people, but relatively young,” he recalled. “I had to learn that. They did not.”
Rachel Happe, founder of the Community Roundtable, reinforced the critical challenge of non-determinism: “With spreadsheets, you still get to a data point that’s fixed. AI is pulling in stuff from everywhere.” With spreadsheets, any calculation traces back to its source. With large language models, provenance remains opaque — making it harder to develop the questioning instinct expertise requires.
The use of generative AI in King’s own research firm illustrates both sides of the paradox. AI has compressed secondary research from nearly a week to a few hours. But “half to three-quarters of that is fact-checking and confirming that the information is correct.” The efficiency gain is real, but it depends on having people with enough foundational knowledge to catch errors.
The implication: accelerating production through AI may inadvertently starve the pipeline of future experts capable of exercising the human judgment that remains essential.
The Anthropomorphization Problem
The Fellows grappled with how AI’s tone and personality affect perception and decision-making. Schwarz noted that “30% of your communication is tone” (with 7% words and 63% body language), and AI outputs carry tone that humans respond to regardless of knowing the source.
The challenge: compelling presentation of synthesized information may lead users to accept insights uncritically without recognizing underlying sources or considering contradictory evidence.
Predictions for 2026 and Beyond
Crossing the Chasm
Holtz predicted that 2026 will be the year generative AI adoption “crosses the chasm” into mainstream adoption.
The mechanism: AI integration will become seamless within existing tools rather than requiring separate applications. Users will encounter AI capabilities embedded in familiar software, reducing friction and normalizing interaction.
This prediction aligns with enterprise investment trends. Deloitte predicts that 25% of enterprises using generative AI will deploy AI agents in 2025, growing to 50% by 2027 (Deloitte, 2024). Meanwhile, 78% of organizations plan to increase their AI spending in the next fiscal year (Deloitte, 2025). EY’s research shows organizations allocating more than 5% of IT budget to AI achieve 70-75% positive project outcomes, compared to only 50-55% for minimal spenders (EY, 2025).
The Agentic Layer
Wu predicts increased development of multi-agent AI systems: “2025 is the year of agents, and next year is where we’re going to actually use these AI to discover new science — fundamental science that has compounding effects to society.”
The promise: once agentic systems can autonomously research, hypothesize, and iterate experiments, breakthroughs in biology, energy, and materials science will accelerate dramatically. The caveat: “Even the communication protocols between the agents are not fully standardized, so the agentic layer came a little slower than we think.”
Progress is measurable but behind expectations. Current research bears this out: 23% of organizations report scaling an agentic AI system somewhere in their enterprises, with an additional 39% beginning experimentation with AI agents (McKinsey, 2025). However, most organizations that are scaling agents say they’re only doing so in one or two functions, and adoption faces substantial barriers including data accessibility and quality gaps, cultural apprehension, and organizational inertia (McKinsey, 2025).
The technical challenges of coordinating multiple AI agents with different perspectives — essentially creating AI teams — mirror human team coordination challenges, suggesting the principles of team flow may apply even in AI-to-AI collaboration.
Secondary Innovation
Wu emphasized the importance of “secondary innovation” — use cases that depend on previous innovations and cannot be envisioned until those foundations exist. “There’s no way you could envision something like Uber or Lyft before you have a mobile phone,” he noted. The implication: genuinely novel applications of AI may remain invisible until the agentic layer matures. “Whatever takes place is really what physics, science, and technology would allow. But if it doesn’t violate any fundamental laws in science, maybe it’s all possible.”
The Corporate Ethics Question
McClure offered a more aspirational prediction: “My hope is that in our post-regulatory environment, particularly in the US, companies and boards will realize their crucial role in technology ethics.” This tension between AI’s potential for human flourishing and its potential for vast economic impact, both potentially positive and negative, may define the next decade of AI adoption.
This ethical imperative becomes even more urgent in light of the AI oversight and governance void at the corporate board level. As NACD notes, boards now stand at the inflection point of transitioning from AI education and awareness to more strategic and integrated AI governance within board operations (NACD, 2025). However, engaging in strategic AI discussions remains challenging when, according to Deloitte’s 2025 Governance of AI report, 66% of boards have limited to no knowledge or experience with AI, and 31% of boards say AI is not on the board agenda.
“My hope and prediction,” said McClure “is that in 2026 boards will recognize that in order to effectively govern during through this time of technological transformation, they must keep pace with AI adoption by embedding AI considerations into their oversight frameworks through board composition that includes AI knowledge and expertise, revised committee charters that prioritize AI governance, regular strategic discussions, and clear metrics to measure the value, risks and opportunities of AI adoption.”
The Singularity Is Human Adaptation
Perhaps the most profound insight came from Wu’s reframing of the singularity concept: “Singularity is not about AI trying to destroy humans. It’s the point where innovation happens so quickly that we as a society, as a race, as humanity, fail to adapt.”
This shifts the risk from machines to human systems. The danger isn’t AI achieving consciousness and turning hostile — it’s the pace of change exceeding our institutional, cultural, and psychological adaptation capacity. “When change happens faster than culture, institutions, and psychology can integrate them, systems break,” as Heuer summarized. “We’ve been going too fast for too long for humanity to keep pace. What happens to our humanity when we go exponentially faster?”
Schwarz added perspective: “The question isn’t whether change occurs, but whether intentional human judgment shapes the trajectory or adoption proceeds by inertia and economic pressure.”
Conclusion: The Human Work of AI Adoption
Three years into the generative AI era, the Fellows’ conversation reveals a landscape defined more by human factors than technical limitations. The technology can compress months of work into weeks, unlock individual flow states, and occasionally produce genuinely novel solutions to problems that have stymied human experts. Yet widespread adoption remains constrained by emotional resistance, craft identity, non-deterministic outputs, and the absence of clear enterprise strategies and robust technology governance and enablement frameworks.
The data confirms what the Fellows observed: a critical gap between AI capability and organizational readiness. While nearly all organizations (99%) have integrated or are integrating AI into their initiatives, fewer than half (50%) are actively investing in governance frameworks to address the risks of emerging AI technologies (EY, 2025). McKinsey’s research shows that only 21% of organizations have fundamentally redesigned workflows as a result of generative AI deployment — the single strongest driver of business impact (McKinsey, 2025). Meanwhile, Deloitte’s surveys reveal that resistance to adopting generative AI solutions frequently stems from unfamiliarity with the technologies or skill and technical gaps, slowing project timelines (Deloitte, 2024).
The path forward requires:
Fostering organizational readiness through clear alignment with strategy, manager encouragement to improve employees’ emotional readiness, visible champions, and cultures that embrace experimentation over fear.
Reframing AI’s role from its focus on automation to augmentation, focusing on how it enables human flow, creativity, and judgment rather than replacing human work.
Solving the determinism problem to enable trust-based deployment in high-stakes applications without requiring 100% validation.
Developing new coordination protocols that preserve collaborative sense-making while accelerating production work.
Maintaining human judgment at the center of accountability, strategic direction, and the essential question of what problems are worth solving.
Building governance structures that match the pace of adoption.
The organizations that successfully navigate this transition will recognize that human innovation, framing, contextualization, and judgment remain irreplaceable. What has changed is the speed at which we can iterate between question and answer, insight and implementation, prototype and product.
As we move into 2026, the question facing every organization isn’t whether AI will transform work — it already has. The question is whether that transformation will be shaped by intentional design aligned with human flourishing, or by the inertia of efficiency optimization and economic pressure. The answer will determine not just the future of work, but the essence of what we choose to value as human.
About the Team Flow Institute
The Team Flow Institute is dedicated to shaping a human-centric future of work as we face the choice of augmentation or automation in every industry and every function. Our mission is to gather like-minded individuals and organizations to steer our collective destiny toward a more sustainable future, where the essence of humanity and human work is valued and preserved as we increasingly adopt AI tools and technologies.
Our Research Fellows bring deep expertise across business leadership, organizational design, technology strategy, and human performance to advance the science and real-world application of team flow principles
References
Deloitte. (2024). State of Generative AI in the Enterprise 2024. Retrieved from https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-generative-ai-in-enterprise.html
Deloitte. (2025). AI Trends: Adoption Barriers and Updated Predictions. Retrieved from https://www.deloitte.com/us/en/services/consulting/blogs/ai-adoption-challenges-ai-trends.html
Deloitte Global. (2024). 2025 Predictions Report: Generative AI. Retrieved from https://www.deloitte.com/global/en/about/press-room/deloitte-globals-2025-predictions-report.html
EY. (2025). Responsible AI Pulse Survey 2025. Retrieved from https://www.ey.com/en_sg/newsroom/2025/07/ai-adoption-outpaces-governance-as-risk-awareness-among-the-c-suite-remains-low
McKinsey & Company. (2024). Most Businesses Lack a Clear AI Adoption Roadmap. Retrieved from https://www.ciodive.com/news/enterprise-lack-generative-ai-strategy-McKinsey/717865/
McKinsey & Company. (2025). The State of AI in 2025: Agents, Innovation, and Transformation. Retrieved from https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
McKinsey & Company. (2025). Insights on Responsible AI from the Global AI Trust Maturity Survey. Retrieved from https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/tech-forward/insights-on-responsible-ai-from-the-global-ai-trust-maturity-survey
National Association of Corporate Directors (NACD). (2024). Tuning Corporate Governance for AI Adoption: 2025 Governance Outlook. Retrieved from https://www.nacdonline.org/all-governance/governance-resources/governance-research/outlook-and-challenges/2025-governance-outlook/tuning-corporate-governance-for-ai-adoption/
National Association of Corporate Directors (NACD). (2025). 2025 Public Company Board Practices and Oversight Survey: AI Analysis. Retrieved from https://www.nacdonline.org/all-governance/governance-resources/governance-surveys/surveys-benchmarking/2025-public-company-board-practices–oversight-survey/2025-board-practices-oversight-ai/
This report synthesizes insights from two Team Flow Institute Research Fellows roundtable discussions held in November 2025.
Editor & Contributors
Editors: Jennifer McClure, Senior Fellow & Chris Heuer, Managing Director
Contributors:
Chris Heuer, Co-founder
Jaime Schwarz, Co-founder
Jennifer McClure, Senior Fellow & Advisor
Rachel Happe, Fellow
Shel Holtz, Fellow
Steve King, Fellow
Dr. Michael Wu, Fellow
Note: This paper was written with the help of Claude.ai’s generative AI technology to aid with editing and some content creation. Any AI-generated content has been reviewed, adjusted and enhanced with experience and insights by a human editor to ensure accuracy, relevance and authenticity.
