Humans and AI in the Feedback Loop of Software Development

The main challenge of software development is the immense complexity of the systems we build. This complexity can only be managed through continuous feedback.

Every change generates feedback: through unit tests, continuous delivery, monitoring, or from users. These feedback loops help teams evolve and improve their systems incrementally.

Software development thus becomes an interplay between feedback systems and human decision-making. Humans interpret the feedback and determine the next steps, making them a crucial part of the control loop in modern, automated software development.

This talk explores how these insights can be used to optimize software development and what role AI can play in that process.

Transcript

1. Humans and AI in the Feedback Loop of Software Book an Development appointment Eberhard Wolff Head of Architecture https://swaglab.rocks/ https://ewolff.com/ with me!
2. What is Software Development?
3. Software
4. Software
5. Software User, Manager, Customer, other Stakeholders….
6. Software Developers, architects, … User, Manager, Customer, other Stakeholders….
7. Productivity Improvement & Wor • AI: not the first productivity improvement • Despite productivity improvements, more and more people work on software. • Why?
8. Rebound Effect • Fuel-efficient car makes driving cheaper. • People drive more • Fuel consumption grows. • Rebound effect: More efficiency -> more total demand
9. Watch, Stopwatch, Calendar Health, fitness, GPS, maps
10. Development Teams • Software development: Team effort since the fifties • No productivity gain changed that. • AI won’t replace teams with a single human. • Science fiction anyhow: Current productivity gains much too low.
11. How Can We Succeed in Software Development?
12. Development = Communication • Asked technical people: Most important skill? • Communication • Main challenge in architecture? • Mostly soft skill related https://software-architektur.tv/2024/08/16/was-ist-der-wichtigste-skill-in-der-it.html https://software-architektur.tv/2025/06/27/was-ist-die-hauptherausforderung-der-software-architektur.html
13. Development is about Communication • How do developers spend their time? • 58% time on development-heavy activities • 15% coding • 36% on collaboration-heavy activities • 15% meetings https://www.microsoft.com/en-us/research/wp-content/uploads/2019/04/devtime-preprint-TSE19.pd
14. An hour of meetings can save hours of coding.
15. How Is Software Development Different from other Activities?
16. Is Software Engineering Engineering ? • Hillel Wayne interviewed “crossovers”. • I.e. software engineers that used to be engineers in other fields • Consensus: Software engineering is engineering https://software-architektur.tv/2024/03/27/folge209.html
17. Is Software Engineering Engineering ? • All engineering aims at iteration. • But the price of an iteration is different. https://software-architektur.tv/2024/03/27/folge209.html
18. Porsche 959
19. Porsche 959
20. Website:Cheaperand Quicker
21. Inner / Outer Loop Release Feature Build Code Outer Loop Inner Deploy Loop Test Planning Customer Feedback
22. Inner / Outer Loop Software is soft – we are Release Feature already quick! Build Code Outer Loop Inner Deploy Loop Test AI might help Planning Customer Feedback
23. DORA: Lean Optimization Idea Production Increase Speed! Bottleneck becomes obvious Testing too slow! https://dora.dev/ https://software-architektur.tv/2020/08/14/warum-continuous-delivery-die-devops-studie.html
24. DORA: Lean Optimization Idea Production Increase Speed! Eliminate Bottleneck https://dora.dev/ https://software-architektur.tv/2020/08/14/warum-continuous-delivery-die-devops-studie.html
25. DORA: Lean Optimization Idea Production Increase Speed! Next Bottleneck becomes obvious Deployment too slow! https://dora.dev/ https://software-architektur.tv/2020/08/14/warum-continuous-delivery-die-devops-studie.html
26. Is AI’s Acceleration Needed? • Every project should deploy as often as possible! • Overwhelming benefits • But: This is not even common knowledge! • Why would projects optimize feedback loops then? • More (AI) speed is not necessarily better feedback.
27. What are Typical Problems in Projects? • Secret ecommerce strategy • Really: no ecommerce strategy • Management decides about architecture • Value and reasoning unclear • PO lets developers fill the backlog • Backlog should be fundamental for POs.
28. What are Typical Problems? • Business value of the whole effort unclear. • Customer / stakeholder / PO not available or not enough time • Excessive planning instead of iterations • This must be done on February 1st 2028! • Really six month sooner or later would be fine. • And actually the scope will be changed anyhow. • Violating fundamental engineering practices!
29. AI Won’t Improve Feedback Release Feature Build Code Outer Loop Inner Deploy Loop Test AI might help Planning Customer Feedback
30. AI: Build the Wrong Stuff Even Faster? Release Feature Build Code Outer Loop Inner Deploy Loop Test AI might help Planning Customer Feedback
31. AI Speed Solves Problems? • AI is quick! • Let’s do 2-4 days development cycles! • How do you solve customer / stakeholder / PO not available? • What about the urge for excessive planning? • Why are we not already doing quicker cycles? • Why not as fast as possible? • I.e. review and deploy each finished story?
32. Can We Fire People because of AI?
33. Big Tech Fires People • Big tech fire people not because of productivity gains… • …but because they need the money for AI data centers.
34. FiringPeople • How many people can be fired before it impacts outcome? • How bloated are organizations?
35. Manager: Prestige Large Team Parkinson: All Do Some Work Communication Fail Conway: Architecture Complex https://www.youtube.com/watch?v=3MP-4UcAYJU https://www.youtube.com/watch?v=p7r6IE7TkpU Conway 1968!
36. Engineers love engineering Complex engineering is more challenging …and provides more prestige. https://www.youtube.com/watch?v=3MP-4UcAYJU https://www.youtube.com/watch?v=p7r6IE7TkpU
37. Many Organizations are too Bloated • Complexity -> prestige for managers & techies • Organizations bloated • Firing people easily possible? • “I fired people because AI!” • The new source of prestige? • Impact job market long-term?
38. AI as a Threat • AI: lasting negative impact on the job market? • Sufficient: believe AI leads to more productivity • Like with off shoring • Frustrating jobs (“review AI slop”) • This might kill our culture.
39. What Will Help?
40. Become an Expert for the Domain and Com • What is specific about your company? • Who should you talk to for a specific development task? • How should you talk to them?
41. Improve Collaboration! • Collaborative Modelling • Post Its • Easy to understand • Fewer barriers • Work in parallel • Easier for introverts to participate
42. Example Collaborative Modeling: Event Stor Explore domain with technical people + domain experts!
43. Socio-technical Architecture Reviews • Architecture reviews often focus on the software. • Software is a product of a social system i.e. a team. • Let’s focus on the team! • Result: Working teams instead of a slide deck
44. https://software-architektur.tv/2026/02/06/folge301.html https://swaglab.rocks/star-review/
45. Conclusion
46. Conclusion • Software development is a team effort. • Communication drives software development. • The real problems are about people. • Become an expert on domain, communication, and people!
47. Thanks!
48. Slides + more Send email to entwicklersummit2026@ewolff.com Powered by Amazon Lambda Email address logged for 14 days, wrong addressed emails handled manually