State of AI4SDLC: how AI changes development
Adoption has happened. Next come systems whose code can burn and regrow
Slide contents
1. State of AI4SDLC: how AI changes development
Adoption has happened. Next come systems whose code can burn and regrow
2. Code is no longer the hard part
3. Four parts of one conversation
Industry: adoption happened, system effect did not
Fintech 10,000+: platform, agents, metrics
Regenerative Software: code meant to die
Engineer: intent, context and verification
4. 01. What happens to the industry
The 2026 cut: Sonar, DX, Harness, Perforce and our AI4SDLC Research
5. Adoption outpaces trust
6. Faster code, flat delivery
7. The bottleneck moves right
8. 02. Inside a large fintech
10,000+ engineers, thousands of services, one control perimeter
9. Three agent platform layers
10. Agents across the SDLC
11. What to count instead of AI-LOC
12. 03. Regenerative Software
Based on Chad Fowler’s book (O’Reilly): what must outlive the implementation?
13. Code that is meant to die
14. The deletion test
15. Seven undeletable primitives
16. Intent: promise, not mechanism
17. Not a single dropped request
18. Architecture is the compile target
19. Framework became architecture
20. Compilation: crisp boundaries
21. Evals check behavior, not code
22. How to verify behavior
23. Provenance: why, not what
24. A decision log instead of memory
25. Pace: layers, different speeds
26. Aerospike: the layer, not the diff
27. Deletion: remove safely
28. 2008: the month would not close
29. Compaction: five levels
30. Compaction at scale
31. Primitives are familiar practices
32. Code remembers, the org forgets
33. Where regeneration breaks
34. 04. How the engineer's work changes
The same platform yields different effects in different teams
35. What set successful teams apart
No effect
Turned on the assistant and waited
Tasks assigned “by chat”
Review and tests unchanged
With effect
Requirements became agent contracts
Review adapted to AI changes
Context and checks in the pipeline
36. Engineer's day: before/after
37. What stays with the human
38. Three skills close the loop
39. Myths not worth fearing
Engineers do not vanish: judgment matters more
Juniors are needed; training changes
Review checks contracts and proof
Accountability stays with the human
40. Five AI4SDLC shifts
Coding is no longer the bottleneck
Queues moved to framing and verification
Knowledge moves into the primitives
Evals and provenance outlive the implementation
Work goes to engineers, not coders
Code burns; systems regrow — after Chad Fowler
41. Thank you!
AI4SDLC
Slides and sources are in the Telegram channel; Chad Fowler's “Regenerative Software” is on O'Reilly
Alexander Polomodov, CTO & Technical Fellow, polomodov.tech
@ai4sdlc