The adrenaline rush is gone and the lights are on. Google Notebook LMs agents dig into Deloitte’s latest State of AI in the Enterprise and confront a tough truth: access exploded, but value is uneven and the governance gap is widening.

Instead of more shiny pilots, 2026 demands systems thinking, economic rigour, and clear decision rights as AI moves from chat to action.

At a Glance / TLDR:

access rising but daily usage lagging
pilot success versus production economics
three tiers from surface gains to deep transformation
revenue gap between savings and new income
job redesign, broken ladder, and pod-based teams
sovereign AI, local models, and data control
agentic AI, tool use, and governance deficits
physical AI growth in APAC and safety needs
2026 as a friction year demanding brakes

The podcast starts with the usage gap - why sanctioned tools sit idle - and trace the roadblocks that turn successful sandboxes into expensive production failures. From latency and cost blowouts to brittle data pipelines, we unpack what it takes to move beyond proof-of-concept purgatory. Then we map the three tiers of adoption: surface-level productivity, process redesign, and deep transformation. A standout case turns mining equipment into connected platforms, shifting from digging to predictable, data-driven extraction. That’s the leap from automation to imagination, and it’s where new revenue lives.

The conversation gets candid on jobs. When models make the call, humans can’t be left as rubber stamps. We explore role redesign, escalation rules, explainability, and the “broken ladder” problem created by automating entry-level tasks. A promising answer is pod-based teams - small cross-functional units orchestrating fleets of AI agents - where learning shifts from manual repetition to supervision and exception handling. We zoom out to sovereign AI and the rise of compact local models that run under domestic rules, balancing control, privacy, and latency with the realities of global operations.

Agentic AI is the tipping point: systems that plan, act, transact, and iterate toward goals. The value compounds, but so does the blast radius of mistakes. With 74 percent planning agents soon and only 21 percent ready on governance, we lay out practical brakes: scoped permissions, human-in-the-loop gates, immutable logs, simulator testing, budget limits, and kill-switches. We also scan physical AI - robots and drones scaling fastest in APAC - where safety and uptime meet AI reliability.

If you’re leading AI adoption, ask three things:

Are we transforming what we sell, not just how we work?
Do we know who overrules the model and when?
And have we built the brakes for autonomy before hitting the gas?

Subscribe, share with a teammate who owns the roadmap, and tell us: what’s the first brake you’ll install?

Support the show

𝗖𝗼𝗻𝘁𝗮𝗰𝘁 my team and I to get business results, not excuses.

☎️ https://calendly.com/kierangilmurray/results-not-excuses
✉️ kieran@gilmurray.co.uk
🌍 www.KieranGilmurray.com
📘 Kieran Gilmurray | LinkedIn
🦉 X / Twitter: https://twitter.com/KieranGilmurray
📽 YouTube: https://www.youtube.com/@KieranGilmurray

📕 Want to learn more about agentic AI then read my new book on Agentic AI and the Future of Work https://tinyurl.com/MyBooksOnAmazonUK