What does it take to build AI responsibly when the technology is evolving faster than the rulebook?

In this episode of The Databricks Diaries, Daniel Thornton sits down with Anna Gressel, Partner and Global Co-Head of AI at Freshfields, to explore the legal, governance and strategic challenges shaping the next generation of artificial intelligence.

From advising some of the world's leading AI developers and global enterprises to helping boards navigate AI risk, Anna offers a unique perspective from the intersection of technology, law and business strategy.

Together, they discuss:

• Why AI governance is no longer just a compliance exercise
• The shift from chatbots to autonomous AI agents—and why it changes everything
• How organisations can balance rapid AI experimentation with responsible governance
• The biggest mistakes companies make when implementing AI
• What leaders should be asking themselves to assess their AI readiness
• Why today's junior employees may soon become managers of teams of AI agents
• The emerging importance of AI incident response and cybersecurity
• How legal teams are becoming strategic partners in AI transformation
• Why "world models" could be the next major breakthrough in artificial intelligence

Whether you're a CIO, CTO, CDO, Head of Data, AI leader or technology executive, this conversation provides practical insights into preparing your organisation for an AI-driven future while staying ahead of regulatory, operational and competitive risks.

If you're thinking about AI adoption, agentic AI, governance, or the future of intelligent systems, this is an episode you won't want to miss.

How do organisations move beyond deploying AI tools and begin creating measurable, sustainable value?

In this episode of Databricks Diaries, Andy Davis speaks with Alexandra Diem, Tribe Lead and SVP for Data, Analytics and AI at Gjensidige, about the practical realities of implementing AI within a large, regulated organisation.

Alexandra explains how her team began by using generative AI to remove tedious and repetitive work for analysts. However, the project quickly exposed a wider challenge: AI is only as useful as the data, metadata and documentation supporting it.

They discuss why data governance, join logic, business definitions and semantic layers are becoming increasingly important as organisations introduce natural-language access to data.

The conversation also explores the changing economics of AI. As developers become more productive with tools such as GitHub Copilot, organisations must consider model selection, token consumption and FinOps practices to ensure that increased productivity does not create uncontrolled costs.

Andy and Alexandra also examine the difficulty of proving the value of internally focused AI initiatives, where the connection between investment and revenue is often less visible than it is with customer-facing products.

Topics covered include:

• Identifying genuine organisational problems before selecting an AI solution

• Using AI to automate repetitive analytical work

• Why metadata and documentation are critical for trustworthy AI

• Building automated governance into the data platform

• Managing AI consumption, model selection and token costs

• Balancing short-term business cases with longer-term AI investment

• Creating effective semantic layers for people and AI

• Why back-office processes may offer the greatest immediate AI opportunity

• Helping developers use AI productively without becoming overly dependent on it

Alexandra’s central advice is simple: do not begin by asking what you can automate with AI. Begin by identifying where users are getting stuck, where processes slow down and where handovers create friction. Those bottlenecks are often where AI can generate the most meaningful value.

Databricks Diaries is hosted by Andy Davis and explores how organisations are using data, analytics and AI to deliver practical business outcomes.

In this episode of Databricks Diaries, Andy Davis is joined by Sam Worthington, Chief Data Officer at Crux Underwriting, for a practical conversation on how insurance organisations can get real value from AI.

Sam shares his perspective on where AI is most useful in insurance, particularly around data ingestion, extraction, underwriting augmentation, bordereaux processing, fraud detection, operational efficiency and decision support.

The conversation explores why AI is not a shortcut around poor data foundations, and why insurers need to think carefully about trust, confidence scoring, conflicting data sources and when human expertise still needs to sit in the loop.

Sam also explains why London Market insurance presents a different challenge to more standardised personal lines environments. When risks are complex, specialist and often difficult to compare, AI needs to support underwriting expertise rather than simply replace it.

Key topics include:

• Why AI value in insurance often comes down to efficiency and decision support
• The role of AI in underwriting augmentation
• How insurers can use AI for ingestion, extraction and cleaner data capture
• Why speed to quote matters for MGAs
• The importance of data foundations before applying AI
• How to manage conflicting insurance data, such as slips, emails and submissions
• Why human referral and confidence scoring are critical in underwriting workflows
• The difference between top-down AI use cases and bottom-up agent adoption
• Why London Market insurance is “lumpy and bitty” compared with more standardised markets
• How to identify the right AI use cases by starting with real business pain points

Sam’s advice is clear: start with a genuine problem, embed the solution into the way people already work, build trust through controls and guardrails, and avoid trying to solve everything at once.

A valuable listen for anyone working in insurance, data, underwriting, analytics or AI who is trying to separate practical opportunity from AI noise.