AI-Powered Domain-Driven Design

LLMs are powerful, but they are not very sophisticated in managing business language. Their very nature can accelerate the software development trajectory towards Big Balls of Mud. We are going faster, but in the wrong direction.
Luckily, Domain-Driven Design offers useful constructs for building our harnesses and supporting business-oriented agentic development.
At the same time, Domain-Driven Design emphasizes learning and mastering domain complexity, and this is where the AI contribution must be managed wisely.

Transcript

1. Ai-powered Domain-Driven Design Alberto Brandolini
2. Disclaimer
3. Photo by Matt Palmer on Unsplash
5. A picture of three controversial heads of state That I removed, to check whether this was activating any filter
7. Domain-Driven Design AI
8. Living the dream
9. Durable results? Photo by Phillip Flores on Unsplash
10. Photo by Grant Hedges on Unsplash Let’s vibe!
11. Context Small Projects Low risk, low harm Based on personal needs -> Domain Expertise
12. What did I learn?
13. AI wasn’t trained like we did
14. AI and US Our skills aren’t evenly distributed Domain Models AI won’t beat you in what you’re really good at. Framed for disappointment AI will help you A LOT in what you’re not that good at Unfortunately, this is also where you’ll be fooled. Frontend development Regexp
15. Where should we invest our time?
16. Curiosity and strategy may pull in different directions
17. Lessons from Vibe Coding Amazing prototyping speed Integrations almost for free! ❤ Genie approach: “You only have three Desires” You have to think like a developer You don’t want to see the code.
18. Photo by Nazym J. on Unsplash Non-Trivial Greenfield
19. Context Post Vibe Coding (🗑🗑😒😏) -> Harness Engineering In-House Exploration Solo Mode One Legacy Application (15 Bounded Contexts) Two Main Scenarios: Rewrite / Porting (Ruby to Java) Refactoring -> later.
20. Business Stakeholder Software Developer Domain Expert Software Architect Entrepreneur DDD Language Freak Photo by Héctor J. Rivas on Unsplash
21. Training Set Hypothesis Average Code & Reinforcement Learning might not be enough Code Good DDD Code I actually like Good DDD Code Wannabe DDD Code Good Code
22. First Experiment Provide Robust Domain Knowledge Escaping the prompt -> Harness Establish a Reference Glossary Describe the desired Architecture(s) setup
23. Imitation beats Guessing
24. Greenfield approach Turn it into a clean brownfield Example features Reference implementation Add features incrementally Inside a coherent harness
25. Progressive Disclosure Agents.md Agents.md A single markdown file is token-heavy and progressively inefficient A structured tree of ondemand information keeps the Token usage under control Frontmatter to fine-tune loading info TopicA.md TopicB.md TopicC.md
26. Mapping Domain Knowledge Domain-Vision.md BoundedContext-A.md BoundedContext-B.md ACL-X.md ACL-Y.md Same principle, but applied to domain knowledge
27. Vision Summary An AI-Readable short document, containing: A short description of the business scenario, A list of the known Bounded Contexts, with a single paragraph description. Domain-Vision.md
28. Bounded Context Descriptor BoundedContext-A.md A single file per Bounded Context Domain Narrative Key features Scenarios Integrations Glossary Banned Terms Our canvas may help! -> https://www.avanscoperta.it/en/bounded-context/
29. CQRS-ES.md Architectural Styles Multiple Bounded Contexts can have the same architectural approach Externalize with examples and common practices
30. ACL Descriptor Business Rationale History: “We used this tool till 2023, then …” Translation logic Local Glossary Disclaimer: really “fresh” ACL-X.md
31. Bounded Context matching Context Window
32. Domain-Aware Harness Probably the single most effective move Convergence on the official narrative “Plausible” test cases appeared naturally The initial context matters
33. A durable recipe? Photo by Phillip Flores on Unsplash
34. Where should they live? Markdown files in GIT? Only the development team knows about it Wiki pages? Non-developers can alter the source of truth Read-only Wiki pages? Giving a weird message…
35. Domain Harness is not a software-only artefact
36. Coherence matters
37. Accessibility matters
38. Enforcing Coherence Matters A non-contradicting harness speeds up operations AI generates redundancy for free Be prepared for some pruning work
39. Can we start from this?
40. Well… more of • Seamless integration - WOW! • Disappointing results • Optimizing felt like the wrong thing to do
41. EventStorming to AI Technically doable Missing details EventStorming is a draft model, not a specification Digital format is already a loss Summarizing Matters There are better formats for consumers
42. Human Brain Are we nuts? Photo by Priyanka Singh on Unsplash
43. Reinventing Flows
44. BDD as a discovery tool Given an open sale for event “X” with a capacity of 200 seats When we register a sale for 5 tickets Then the resulting capacity should be 195 seats
45. BDD as a validation tool Given an open sale for event “X” with a capacity of 200 seats And we agreed with company ABC to resell 80 tickets And we sold 140 tickets overall Then the resulting capacity should be 60 seats And Company ABC could not offer more than 60 seats Did we understand correctly?
46. Software development is a learning process Working code is a side effect https://blog.avanscoperta.it/2026/08/27/code-is-still-a-side-effect/
47. AI is blind to any nondigital artefact Selective Blindness Photo by stephan sorkin on Unsplash
48. AI Can’t see non-digital tools Or uncommitted code
49. AI Can’t see Learning Only learning outcomes
50. Conversations Drafted Acceptance Tests Workshops Sketches Speech Prompt AI generation AI Plans Code gen era er n e g AI tio n n o i t a Presentations Digital Images Docs A g I a r e en n o ti
51. BDD is not what it used to be The business validation is not part of the flow 😒 Less interactive sharpening 😒 A different test runner, trained the wrong way. 😔 You may want to have a look to: https://varar.dev/
52. TDD is not what it used to be Fast support to complete coverage. 😏 No design friction. 🤔 No Red/Green/Refactor. Hiding technical debt under test coverage. 😳
53. How the brain works Language inconsistencies Learning Context Le a Translation AI Harness Information availability rn Models in Politics g g n i n r a Le Token efficiency Harness
54. Optimize for learning Our brain hasn’t changed, yet Hands-on, Lo-fi activities are still more brain-friendly Shortcuts are not helping learning Summarizing is (painful) learning too
55. Attrition is learning
56. How can AI help? Covering areas where we have no expertise Challenging the current assumptions Analyzing competitors …
57. Grill me?
58. Grill-me & Friends Pull towards a waterfallish approach “Give me all the details NOW, so that I can do my magic in one shot” Where do we get all the details?
59. The Wrong Whirlpool Getting pulled into the wrong media, Ideal for AI, but not for you Photo by Simon Hurry on Unsplash
60. Can we interact better?
61. Media Matters Speech != text Drawings != ASCII Art Interactive Sessions != Bulk execution
62. The whirlpool principles Different perspectives provide more insights A single tool will progressively have decreasing ROI Switch to the tool that provides the most useful feedback!
63. AI offers more tools
64. More speed
65. Structured Flows
66. Land and explore
67. Land and explore Get to a plausible implementation ASAP Refine with more precise details once you have established actionable feedback loops Reinforce with Examples and specifications for corner cases
68. Indulging in details https://blog.avanscoperta.it/2026/09/04/the-sweet-taste-of-indulgence/ Photo by Caroline Ross on Unsplash
69. Structured Flows Pressure for “the perfect flow” Every feature feels different Negleted details can make a difference “Styles” need to be maintained and selected upfront Everything will be obsolete next week. Expand your toolkit, establish the mandatory parts, keep experimenting
70. Photo by Nazym J. on Unsplash Non-Trivial Greenfield
71. Greenfield Businee Scenario Domain-Aware Harness Porting and integrations almost for free Beautiful! And Rewarding …Too many hats
72. Not-so-bad legacy Photo by Steven Cordes on Unsplash
73. Context 10+ years of data Modular monolith, CQRS/ES, well-tested Code Quality (9.9 on CodeScene)
75. What was the bottleneck?
76. Business Stakeholder Software Developer Domain Expert Software Architect Entrepreneur DDD Language Freak Photo by Héctor J. Rivas on Unsplash
77. Autonomy was the bottleneck
78. Code Quality was the enabler
79. Conflicting sources
80. Minion Rebellion
81. Who should we believe? Code says one story; documentation says a different one. AI uses AI-generated sources to back itself up Experiment: Block proliferation Experiment: Every Bounded Context defines its source of truth and its evolution stance Specific skills to monitor /language-drift
82. Coherence is a business problem
83. The dark side
85. Without Touching the code Good exploration practice A new Forming (bad) Habit Code is not dirty Maybe the marketing message is not for you Running on 40+ Years of experience. What happens without it?
86. Do we have a learning path? Photo by Alvin Mahmudov on Unsplash
87. Human in the loop
88. The Accountability problem Con dence Code Accountability Con dence Code fi fi Accountability • We can write more code faster! • Yay! • But coding speed was constrained by confidence. (With exceptions) • AI brings more code with less confidence 🤞 • Accountability followed up code and confidence…
89. The big blind bet
90. The big blind bet
91. Non-neutral tool • AI is a continuous pull into bad habits • Sicophancy • Channel choice • Boredom and multitasking
92. Addictive by design Photo by Ays Be on Unsplash
93. Hello, lizard brain!
94. Engineered addiction Chat interface pulls your brain in the wrong direction You need to be in control Offline exploration (paper, markers, whiteboard, parks) Time-Boxed sessions Safeguards …
95. Wrapping up
96. Stories without context are bullshit
97. What are we optimizing for?
98. Optimize for your learning 🧠 Immersive Modelling Brain-intensive summarizing AI- Friendly Digest
99. Shortcuts don’t really help
100. Learning is still a vital asset
101. Domain-Aware Harness Domain-Vision.md CQRS-ES.md BoundedContext-A.md • Describes our domain • Leverages Bounded Contexts • Durable knowledge for the organization BoundedContext-B.md ACL-X.md ACL-Y.md
102. Domain-Aware Developer • Understands the business domain • Connected to experts and practitioners • Autonomous in making experiments
103. Coherence is not a free meal
104. One style won’t be enough
105. Trust me, It’s going to be great!
106. Thank You!
107. Contacts: • Blog: • https://blog.avanscoperta.it • https://medium.com/@ziobrando • https://ziobrando.substack.com/ • BlueSky: @ziobrando.bsky.social • Consulting & Training: info@avanscoperta.it • https://eventstorming.com • http://www.avanscoperta.it