Blogmark: Organisational design and Team Topologies after AI (via)

I'm a massive fan of the book Team Topologies and looking forward to the 2nd edition. I listened to the authors talk on this podcast. Here's some bits that stood out to me:

Technology As Empowerment: Evaluate AI by whether it enables teams to achieve goals, not by novelty or adoption metrics.

Don't optimise locally without understanding the bigger picture: Assess downstream capacity and maintainability before accelerating code generation. Remember that teams are related to each other in a system or organisation.

Generative AI Powers Product Development: Generative AI can generate many options (text, images, product ideas, versions of a service) to test against synthetic users to quickly arrive at an A/B test.

Team Topologies As Infrastructure For Agency: Team Topologies, when originally written, was about trying to provide agency, empowerment, autonomy to teams of humans within an organisation. The principles and patterns behind Team Topologies apply to agents too... respect cognitive limits (context), make changes small and safe, aligned to a business domain. Perhaps even the three team interaction modes could be a way of thinking about agent-to-agent interaction via MCP?

Prepare Foundations Before Scaling AI: The capabilities that predict high performing software delivery - https://dora.dev/capabilities/ - plus work organised around business domains (think DDD). Decoupled, well-defined boundaries, with solid engineering practices in place, will position organisations to take advantage of AI agents and work in parallel. Without these foundations, which is already true for human-only teams, delivery of value will tend to be poorer.

Train Teams To Be Risks Assessors: People should know when to worry about risks that AI presents. Building prototypes the risk is lower, whereas building for production comes with much higher risk, so everyone should be asking what could go wrong, whether that's in the code an agent has built or an AI tool that is deployed. Teams should already be employing techniques like threat modelling anyway so continue to employ this type of practice.