AI Is Changing Software Development. What Matters for Engineers Now? (Series #Engineering)

I spoke at DotNext today with a keynote, where I took Chad Fowler's idea and the book "Regenerative Software" (which I wrote about earlier) and explained how the industry changes when code is no longer where the system's ground truth lives. Naturally, I started with three slides on the state of the industry and three slides on large-scale fintech, but then there were plenty of interesting ideas and many stories from practice. I will gladly share the video later; for now, you can browse the deck, summarized briefly below.

1️⃣ Writing code faster does not yet mean delivering outcomes faster. AI has already become an everyday tool, but systemic impact does not appear automatically. More code and more PRs create more verification work. Bottlenecks shift toward problem framing, review, and acceptance. The decision about what is worth doing at all—and what should be rejected—becomes especially important.

2️⃣ Impact requires redesigning the entire cycle. Consider a fintech organization with 10,000+ engineers: agents need a shared platform with access to models and tools, policies, sandboxes, and observability. They work with tests, security, incidents, and migrations. Outcomes are measured through accepted changes, waiting time, rework, and delivery reliability.

3️⃣ Knowledge about a system must outlive its code. The core of the talk is Chad Fowler's Regenerative Software. An implementation can be replaced if seven things are preserved outside it: intent, architectural boundaries, behavioral checks, reasons behind decisions, rules for the pace of change, safe removal, and reducing complexity. A simple thought experiment: if the code disappeared, could you restore the system and prove that it works correctly? The answer shows where your understanding actually lives.

4️⃣ Before replacement, extract what only the old system knows. Mature code hides business rules, the consequences of incidents, and non-obvious dependencies—including manual work performed by people. First, those requirements need to be identified and protected with checks. Then the system can be changed incrementally, retaining a rollback path and removing accumulated complexity.

5️⃣ The engineer's work is shifting toward responsibility for the system. Formulate intent, set constraints, assemble context, manage agents, and verify results independently. Do not delegate something entirely when you cannot yet recognize that its output is wrong. The cheaper code becomes, the more valuable engineering judgment is.

#Engineering #Software #AI #Leadership #AI4SDLC #Agents #SystemDesign