AI Writes Code Faster. Why Aren't We Shipping Faster?

The manifesto: https://krisitown.github.io/agentic-s... My linkedIn:   / kristiyan-stoyanov-395310184   Blog post: https://dev.to/kstoyanovai/ai-writes-... An agent can finish the implementation while the change still waits for testing, review, and release. Where does the productivity gain actually go? In this talk, I walk through how I think software delivery needs to change as more of our code is generated by AI. I cover the short feedback loop of specifications and local integration tests, the longer loop with real dependencies, and the human side of trusting the evidence those checks give us. The examples come from experience: the 70-plus-file merge request nobody wants to review, the engineer who still checks everything manually, and the believable API contract an agent invented that my review missed too. For lead engineers and engineering leaders, the practical questions are: What should automation take over? What should human review mean? How much scrutiny does this particular change need? And how do we tell whether we are actually delivering sooner? CHAPTERS 00:00 AI codes faster. Where is the productivity gain? 00:31 Code is only one step 01:30 Six commitments for software delivery 02:16 The short loop: specifications and local tests 04:09 What passing tests actually prove 04:35 Why engineers keep checking manually 06:18 The 70-file merge request 07:06 What should human review do? 08:19 Automation must change the work 08:44 The long loop: real dependencies 10:41 When the agent invents the API contract 11:53 Give the agent real context 13:33 Match scrutiny to consequences 15:18 Throughput, time to production, and rework 17:06 Start with one bounded experiment 18:04 Make this a practice 18:39 What would you change? I'd like to hear about your team: once an implementation is finished, where does it wait longest? What would need to change for you to remove that delay with confidence? #SoftwareEngineering #AI #DevOps