Agentic coding offers faster delivery, but how do we keep moving fast without increasing outages that damage the business and drain engineering capacity? This talk shares the experience of loveholidays (https://www.loveholidays.com/de), where doubling our delivery cadence made reducing incidents and their associated toil a priority. Our hypothesis was that persistent errors and noisy developer feedback made new problems harder to recognise.

We took the approach of role modelling a zero-surprises culture at senior individual contributor levels:
- all surprises shall be known
- all surprises shall be resolved
- no surprise shall be allowed to continue to occur indefinitely

Following a proof of concept within our centre of excellence, we spread the practice through our staff engineer community, building executive and grassroots support.

Using a shared collaboration model, engineers shaped our process and drove adoption. They helped define the measures, established daily improvement habits and investigated beyond their core systems. Following errors across team boundaries brought questions of technical, cognitive and intent debt into focus. Making the work visible encouraged others to help and created opportunities for learning across teams.

Recorded production errors fell substantially, with fewer recorded incidents alongside increased delivery. Engineers reported improved shared context and system understanding. The talk will share quantitative results on engineering outcomes and qualitative experiences of our engineers, alongside the business and engineering benefits observed.

Our conclusion is that, in the age of agentic AI, strong human understanding of our systems remains critical to good system governance. We’ll consider how the approach links to Peggy Storey’s triple debt model: retaining virtuous friction and communication between engineers and teams helps improve shared understanding and maintain a shared model of system design, resisting cognitive and intent debt alongside tackling the technical debt exposed by “surprises”. We’ll share the organisational challenges we overcame during implementation, what we learnt, and how it is possible to move towards a zero-surprises system architecture.