You can write code faster. Can you deliver it faster? (Category Productivity)
It's opening. keynote report DPE Summit Conference 2025 (I'm talking about her. mentioned). It is told by Hans Dockter, CEO of Gradle Inc, who is associated with the theme of build automation and delivery pipeline: he founded Gradle Build Tool, led large enterprise-assemblies and has long worked at the intersection of tooling, DevEx and software delivery.
The main thesis of the report is given in the title - GenAI speeds up code writing, but this does not mean accelerating the delivery of working software, the Author notes that if AI increases the volume of changes to upstream, the downstream system - build, test, compliance, deploy - begins to choke and it looks like a warning to listeners. Hans explains the problem this way. AI makes code faster, more frequent and in larger volumes, but at the same time reduces the depth of human understanding. It provokes larger and riskier changes. This worsens feedback loops and increases the number of failures. As a result, there is more and more friction between "experiment quickly" and "reliably delivered to production."
When it comes to speeches, they are like this.
1Focus not on output but on outcome. It is not important to increase the amount of code generated (LoC, lines of code)The increase in the number of working software that the team actually brings to users
2Bottleneck shifts from coding to delivery system If earlier many teams optimized IDE, code review and code generation, now the competitive advantage will be for those who can accelerate pipeline throughput without loss of quality and without explosive growth in the cost of CI. (continious integrations). That is, the winner is not the one who “writes code faster”, but the one whose Pipeline architecture can withstand the AI load.
3GenAI-ready pipeline isn't just more iron The author talks about specific classes of solutions that need to be improved: treadshuting, intellectual parallelization of launches, predictive testing, universal caching, automation of policy application. In general, it will not work to throw runners - you will have to rebuild the system as a smart conveyor with good diagnosis and risk management.
**4Observability and Quick Key Reason Analysis Become Part of Developer Experience It seems.**We are moving towards the fact that AI coding can become an increasingly “black box” for the developer, and therefore it is important to strengthen not only testing, but also the visibility of signals: test results, static analysis, grouping of files, coverage, diagnosis of the causes of falls.
I also liked the metaphor that the author picked up for the talk - he compared the rate of mutations in organisms and the appearance of cancer with the speed of AI coding and mutations of GenAI models that lead to hallucinations. As a result, the author drew parallels between the number of cells in elephants compared to humans and the work of their mechanisms of correction and protection against mutations – if elephants did not have such protection and they had similar means of self-correcting DNA, they would live an order of magnitude less and die from cancer. The parallel here was with our CI/CD tools, which need to be cool so that we can defend against the non-deterministic nature of GenAI models, which now generate code streams for us:)
In terms of conclusions, I think the following four things should be learned.
- Measure not how much code was written, but how the lead time, failure rate, MTTR and time-to-feedback changed.
- DevEx now includes quality delivery loops, not just editor and SDK convenience
- Platform team becomes even more important: it makes AI adoption safe and cost-effective
- The most mature position for the leader is not to argue “will AI replace engineers”, but to rebuild the system so that the growth of output does not destroy the outcome.
In general, AI has made all these things even more important, as they have been on the agenda of many companies before.
#DevEx #Metrics #DevOps #Engineering #Software #Management #Leadership