AI vendors are shifting from flat-fee pricing to consumption-based billing, and most organizations have no idea what their token spend is actually producing.

Dave Sharrock and Peter Maddison break down what's driving the shift to AI token economics, why the old "$20 per user" budget model is breaking down, and why usage alone is the wrong thing to optimize for. They dig into the pattern showing up across organizations, where a small share of users account for half the token spend, and why chasing that number down misses the real question: what value did that spend create? The conversation covers KPI traps, model selection tradeoffs, and how to build the kind of honest, open culture that lets you actually govern AI spend without punishing your best people.

This week's takeaways:
- Token usage by itself is a bad KPI once your organization has moved past early AI adoption, because it stops measuring exploration and starts driving the wrong behavior.
- The 10% of users driving 50% of the token spend aren't automatically the problem. Some are generating outsized value, and the only way to know is to ask them directly.
- Managing AI cost well means pairing spend visibility and caps with an honest conversation about the value that spend is producing, not just sorting a table by usage.

Listen to the full episode at definitelymaybeagile.com
Subscribe so you never miss an episode.
Have a question or topic you'd like us to cover? Reach out at feedback@definitelymaybeagile.com

Giving an AI agent real autonomy means accepting it will fail early and often before it gets good, the same curve organizations hit during any real change.

Peter Maddison brings a stuck OAuth problem to the table: an AI agent that kept going in circles and couldn't find its way through. That leads into a conversation from Dave Sharrock's local AI meetup about an AlphaGo-style approach to AI agent autonomy: instead of specifying every step, you define hard constraints and let the model work out its own strategy inside them. Peter and Dave connect this to the Virginia Satir change curve, the same dip in performance that shows up when an organization tries a new way of working, and to the difference between using AI to optimize what you already do versus using it to rethink the business itself. They also get into how experiments like Andon Labs' AI-run cafes and vending machines use small dollar constraints to let a model learn from failure without real financial risk.

This week's takeaways:
- A well-articulated objective with clear guardrails lets an AI agent find its own path to a solution, even one you didn't expect or fully understand.
- Real learning, whether it's an AI agent or an organization adopting a new way of working, comes with an unavoidable dip in performance that can't be planned away.
- The bigger opportunity with AI isn't squeezing more efficiency out of an existing process, it's using AI to test entirely different ways a business could operate.

Listen to the full episode at definitelymaybeagile.com
Subscribe so you never miss an episode.
Have a question or topic you'd like us to cover? Reach out at feedback@definitelymaybeagile.com

AI can write code faster than ever, but Peter Maddison and Dave Sharrock argue that coding was never the actual bottleneck in software delivery.

In this conversation about AI software delivery, Peter and Dave dig into why faster coding hasn't solved the two problems that always mattered: knowing whether what you built is actually valuable, and knowing what to build in the first place. They connect this back to sprint length, arguing it was never set by how hard the coding is, but by how fast an organization can learn and decide. As AI generates more options and even makes decisions on our behalf, the conversation turns to what happens when judgment can't keep pace with output, and why product owners and stakeholders still need real time to validate high-risk calls.

This week's takeaways:
- Coding speed was never the real constraint. The two problems that still matter are knowing if something is valuable and knowing what to build in the first place.
- Sprint length should be set by your organization's decision and learning latency, not by how fast code can be written.
- As AI generates more options and even makes decisions for you, leaders need time and context to validate high-risk calls, because judgment doesn't speed up as easily as output does.

Listen to the full episode at definitelymaybeagile.com
Subscribe so you never miss an episode.
Have a question or topic you'd like us to cover? Reach out at feedback@definitelymaybeagile.com