Everyone is talking about AI. Almost nobody is telling the truth about it.

Lee Crossley builds tools whose main job is to demonstrate that the thing everyone is buying does not do what they were told it does. He spent years in hardware, ran technical quality for roughly 2,500 engineers across 13 offices, and today writes the audio drivers himself for a product he is building with one other person. His argument in this episode is that AI models have plateaued into a commodity, and that every decision still worth arguing about now sits in the layers above them.

What You Will Learn

• Why AI text watermarking collapses the moment someone paraphrases the output, and what a real provenance trail would have to record instead
• The Monday morning question every CIO should put to an AI vendor, which exposes in one answer whether a rollout is real or theatre
• How to measure an AI productivity claim so that it survives a conversation with your CFO
• Why only 5 percent of enterprise AI pilots produce measurable business impact, and why the research points at learning rather than model quality or regulation
• What a memory layer actually is, why providers over-send your context, and what that costs you in privacy and tokens
• How local speech to text closed the accuracy gap with the cloud, and why that changes what regulated teams are allowed to build
• What Lee looks for when hiring now that CVs are written by AI and screened by AI

About the Guest

Lee Crossley is a consultant CTO, a director at San Digital and co-founder of Overshow, a privacy-first work memory for macOS that keeps screen context, meeting audio and notes on the user's own machine. He was previously Chief Engineer at BJSS and CTO of a hardware startup that took him to Taiwan around 20 times in three years. He is lead author of Production-Grade AI: From Boardroom to Build and Beyond, and he teaches AI-assisted engineering and AI-assisted knowledge work through San Digital.

The conversation with Dr. Sebastian Wernicke delves into the challenges of being data-driven versus data-inspired, the role of culture in decision-making, and the limitations of dashboards in reflecting the true state of a business. Dr. Wernicke emphasizes the need for a shift from data-driven to data-inspired decision-making and the cultural barriers that hinder this shift. The conversation with Dr. Sebastian Wernicke delves into the impact of AI on data culture and the role of data scientists in the age of AI. It emphasizes the importance of purpose-driven AI and the limitations of technology in fixing broken cultures. Dr. Wernicke also discusses the misinterpretation of data and the significance of human skills in data science.

Takeaways

• Data-driven decision-making optimizes existing processes, while data-inspired decision-making seeks to transform and innovate.
• Culture plays a significant role in decision-making, and the use of data is often influenced by cultural factors within an organization. Purpose-driven AI overcomes broken data culture
• Data scientists bring analytical and business skills to the table
• The importance of understanding the human side of data
• The fallacy of managing by numbers

Chapters

• 00:00 Data-Driven vs. Data-Inspired Decision-Making
• 02:21 The Journey from Bioinformatics to TED Stage
• 11:47 The Purpose and Limitations of Dashboards
• 19:53 Culture as a Barrier to Data-Inspired Decision-Making
• 28:29 AI and Data Culture
• 29:02 The Role of Data Scientists in the Age of AI
• 30:12 The Misinterpretation of Data
• 34:05 Human Skills in Data Science
• 44:18 The Fallacy of Managing by Numbers

The conversation delves into AI governance in regulated industries, the gap between what governance policies say and what employees actually do, the human-in-the-loop concept as it works in practice rather than on a process map, and where accountability sits when an AI writes something and a human approves it in four seconds. Daanyaal Bandukwala brings the pharmaceutical commercialization and patient support program lens, and the episode is upfront about the commercial relationship between host and guest, which is why the questioning goes harder than usual. Topics include the Globe and Mail test as a governance heuristic, automation bias in expert decision making, AI disclosure, data residency versus sovereignty, the collapse of the build case in favour of buying, and an honest audit of which governance studies are vendor funded.

Takeaways

• Governance is not killing AI projects. Unanswered ownership questions are. Compliance is asking who is accountable, and nobody wrote it down.
• The Globe and Mail test is the simplest governance heuristic available. If this ended up on the front page tomorrow, how would you look?
• AI governance lives in onboarding, not in a paragraph buried in a large SOP.
• Human in the loop is not the same as human accountability. When AI is wrong, expert accuracy collapses, and only personal accountability changes that.
• A quarterly spot check beats an annual formal audit, and a vendor's reaction to being spot checked tells you more than the audit clause.
• No independent, non-vendor study isolates governance as the cause of better AI outcomes. It is conviction, not proof, and both host and guest say so.
• Start with the business problem, not the platform. Evaluate governance before features.
• Walk out if a vendor claims 100 percent accuracy, says governance can wait, calls integration phase two, or says AI works the same in any industry.
• Executives are wrong about how their people feel and wrong about what their people are doing, and most are comfortable with unapproved use anyway.

Chapters

• 00:00 The Role of Governance in AI
• 03:20 Challenges in AI Adoption
• 19:43 Human-in-the-Loop in AI Practice
• 27:55 Accountability in AI
• 30:48 Responsible AI and Transparency
• 31:58 AI Disclosure and Transparency
• 33:13 AI in Medical Education
• 35:18 Influencer Marketing and AI
• 36:37 AI in Healthcare and Patient Support
• 41:20 Data Resiliency and Sovereignty
• 45:12 The Review Step That Stopped Happening
• 50:07 Regulatory Decisions and Drug Discovery
• 53:11 Build vs. Buy in AI Solutions
• 57:48 Specialized AI Vendors
• 58:30 Why There Is No Good Data on Build vs. Buy
• 1:02:11 How to Choose an AI Vendor
• 1:07:16 Vendor Red Flags and When to Walk Out
• 1:11:52 The Hard Question: Is Governance Actually Proven?
• 1:15:28 Five Beliefs, True or False
• 1:19:18 Shadow AI and the Executive Perception Gap
• 1:23:17 The 95 Percent Pilot Failure Myth
• 1:26:57 Whose Job Gets Smaller
• 1:29:18 Middle Management and Job Security
• 1:32:57 The First 90 Days
• 1:36:01 What Daanyaal Is Least Sure About
• 1:39:57 The Guest Question and Closing

Mykle McKiernan is an enterprise technology leader who has built and run platforms at the highest level, with senior roles at Wayfair and Amazon and earlier work at McKinsey. He has lived through four enterprise technology revolutions from the inside: spreadsheets, ERP, CRM, and now AI. That vantage point is rare, and it makes him one of the few people who can tell you whether this moment is actually unprecedented or just the newest wave.

This conversation traces the impact of AI on enterprise technology by lining it up against the revolutions that came before it. Spreadsheets, ERP, and CRM each arrived with the same fear, the same resistance, and the same eventual payoff, and each one teaches something about what makes technology stick. The throughline is simple: adoption succeeds when people gain leverage and fails when they only see compliance, and value has to reach every stakeholder to hold.

From there the discussion turns to what makes this wave different. Spreadsheets democratized analysis, ERP standardized process, CRM standardized relationships, and AI democratizes intelligence itself. That is why the reaction is so much stronger this time. It is no longer about process. It is about identity. The winners will not be the people who compete with AI. They will be the ones who learn to orchestrate it, who understand that real value shows up quietly years later rather than in the demo, and who are willing to let AI into the work and the hobbies they care about most.

Takeaways

• Technology adoption fails when users only see compliance, it succeeds when users gain leverage.
• Adoption depends on delivering value to multiple stakeholders first and quickly. AI adoption success depends on employee leverage.
• Stakeholders' mutual benefit is crucial for AI adoption.
• The AI wave is different because it is about identity, not process.
• The winners will be those who learn to orchestrate AI.
• AI value shows up quietly years later, not in the demo.
• Embrace AI in hobbies for mutual benefit.

Chapters

• 00:00 The Impact of AI on Enterprise Technology
• 02:02 The Spreadsheet Revolution
• 12:46 The ERP Challenge
• 29:48 The CRM Dilemma
• 34:14 AI Adoption and Experimentation
• 37:27 Stakeholders' Mutual Benefit
• 39:29 AI Wave: Identity vs. Process
• 40:13 Orchestrating AI
• 40:58 Value of AI Shows Up Quietly
• 01:05:29 Embracing AI in Hobbies

The conversation delves into the role of AI as a compression engine, its impact on search behavior, and the loss of critical thinking in research. It also explores the semantic search problem and the feature rank system. The conversation delves into the challenges and implications of AI, including its accountability for misinformation, the challenges in AI adoption and ROI, and the importance of understanding the limitations of AI. It also explores the impact of AI on work and the need to understand when not to rely on AI. The discussion highlights the need for a nuanced understanding of AI's capabilities and limitations, as well as the responsibility and accountability associated with its use.

Takeaways

• AI as a compression engine
• AI's impact on search behavior
• The loss of critical thinking in research AI's accountability for misinformation
• Challenges in AI adoption and ROI
• Understanding the limitations of AI

Chapters

• 00:00 Semantic Search Problem and Feature Rank System
• 32:37 AI and Misinformation
• 34:05 Accountability and Responsibility
• 35:25 The Challenge of Defining Truth
• 38:31 AI Adoption and ROI
• 43:06 False Expectations and Unattainable Goals
• 50:25 AI's Impact on Work
• 59:16 Understanding AI's Limits

The conversation explores the era of agentic AI, understanding AI agents and automation, AI agents as workforce augmentation, and scaling and governance of AI agents. It delves into the challenges, risks, and misconceptions surrounding the implementation and management of AI agents in organizations. The conversation covers a range of topics related to AI strategy, agent development, and the use of language models. It delves into the challenges and considerations for leaders in adopting AI technologies and formulating effective strategies. The discussion emphasizes the importance of staying informed, avoiding vendor lock-in, and strategically leveraging language models in the organizational stack.

Takeaways

• AI agents are not meant to replace human workforce but to augment and enhance it.
• The governance and scaling of AI agents pose significant challenges and risks that need careful consideration and management. Augmented full stack development enables faster agent development
• Caution is needed in buy vs. build approach for agents
• Avoiding vendor lock-in and ensuring portability is crucial for long-term AI strategy

Chapters

• 00:00 The Era of Agentic AI
• 06:56 Understanding AI Agents and Automation
• 18:14 Scaling and Governance of AI Agents
• 27:54 Augmented Full Stack Development
• 36:30 Decoupling from Vendor Lock-in
• 45:12 AI Strategy and Leadership

Disclaimer: The views and opinions expressed by Brent Lewis in this podcast are his own and do not necessarily reflect the official policy or position of Armstrong World Industries, Inc. or any of its affiliates. Brent's participation in this episode is in his personal capacity, and nothing in this conversation should be construed as an official statement, endorsement, or representation on behalf of Armstrong.