Episode 2 – Enterprise AI Field Notes: Traps, Hard Truths, Hallucinations, Shadow AI, and the 3 AI Rooms

Hosts Srini Annamaraju and David Royle are back for episode two – and yes, the feedback is in. Some said we were a bit too serious last time. We’ll try not to become a Sunday love songs show, but we’re working on upping the “gags per minute.”

This week’s conversation covers:

Listener reach & feedback: Almost 100 plays already, with listeners tuning in from the UK, US, India, and even Slovenia. The appetite is real for discussions that go beyond hype and get into the messy middle of enterprise AI.
Event notes from Big Data London: A buzzing show, but still very tech-heavy. We debate whether AI conversations are stuck in the IT lane, and why that’s a problem when the real impact is business-wide.
NBER study on ChatGPT usage: 700M weekly users. Surprisingly, 70% of usage is personal rather than work. Heavy skew toward under-26s. We unpack what that means for adoption inside enterprises.
US tech investment in the UK: Nvidia and OpenAI committing eye-watering sums (hundreds of billions over time). A rare bit of good economic news for the UK, with national implications for jobs, productivity, and independence from US/China dominance.
Enterprise field news:
- Citi experimenting with agentic AI for wealth advisors, using Claude and Gemini inside secure workspaces.
- FT analysis showing CEOs hype AI on earnings calls, but get risk-heavy and muted in regulatory filings.
- JLR cyberattack fallout: £3.5B revenue hit, no cyber insurance in place. Knock-on effects on suppliers and supply chain.
The “three rooms” where AI decisions get made:
- C-suite (value, governance, risk)
- Technical teams (architecture, data quality, safe design)
- Operations (ongoing management, compliance, usage quality)
Traps to avoid:
- The Whac-a-Mole Trap – hallucinations never disappear, they just reduce.
- The Origami Trap – clever prompts aren’t a moat; without guardrails, they fold fast.
- The IT-Only Trap – AI left to technologists will fail; business P&L owners need to lead.
- The Corporate DNA Trap – over-automating risks erasing what makes your org unique.
Shadow AI is real: Even if companies ban AI tools, staff use them on personal devices. Risks around leakage and compliance multiply.

We close with a look ahead:

How frontier model labs (OpenAI, Cohere, Mistral, etc.) are approaching enterprise go-to-market.
Real use cases from our own client work – what’s working, what’s not.

Next steps for listeners:
Got topics you’d like us to cover? Message us on LinkedIn. The more specific, the better – we’ll dig in and bring field notes to the next episode.