Slide contents

1. 1. Why the Architect AI Copilot Still Hasn’t Arrived

   Unpacking a systematic review of 51 studies with Sergey Baranov

2. 2. AI Sees Snapshots; Architecture Keeps History

3. 3. Diagrams Capture Form; Decisions Hold Trade-offs

   Artifact

   Diagram

   ADR

   Pattern list

   Architecture

   Why now

   Cost of change

   Consequences later

4. 4. Review Maps Evidence, Not Copilot Performance

5. 5. 51 Studies Survived Multi-stage Screening

6. 6. Broad Coverage, Shallow Integration

7. 7. Wider Decisions Are Harder to Verify

8. 8. Three Experiments Show Value — and Its Limits

9. 9. LLM Produces Candidates, Not Approved ADRs

10. 10. Runtime Gives AI a Measurable Outcome

11. 11. Results Hold Within Their Measurement Boundary

    What was measured

    Extraction / patterns · 4 projects / 3 classes

    ADR candidate · n=95

    Deployment time · AKS testbed

    What does not follow

    Requirements and trade-offs are covered

    The system boundary is optimal

    The decision survives releases

12. 12. Twenty-one Tools Span Four ISL Levels

13. 13. Tools Create Islands of Intelligence

14. 14. Artifacts Exist. The Architecture Chain Does Not

15. 15. Fifteen Problems Become Six AI Requirements

16. 16. Six Gaps Remain Before a Real Copilot

17. 17. Adaptation Requires Memory of the Long Horizon

    AICH1 · update the recommendation

    A requirement version changed

    Find dependent decisions

    Rebuild the option + evidence

    AICH6 · retain consequences

    Observe debt and smells

    Link versions, incidents, and erosion

    Return the signal to the decision

18. 18. Traceability Without Context Remains Formal

    AICH2 · continuous traceability

    Requirement ↔ ADR · type/version

    ADR ↔ component · rationale

    Component/code ↔ runtime · evidence

    AICH3 · local context

    Domain rules and data

    Teams · regulation · security boundaries

    Migration cost + legacy constraints

19. 19. Expert Review Requires Evidence, Not Confidence

    AICH4 · expert review

    Standards · security · regulation

    Local domain exceptions

    Owner accepts residual risk

    AICH5 · evidence-based measure

    Quality attribute

    Signal + baseline

    Threshold changes the decision

20. 20. ArchBench Standardizes the Pipeline, Not Metric Validity

21. 21. R2ABench Separates Form, Graph, Meaning, and Evidence

22. 22. Knowledge Is Not Decision Ownership

23. 23. Benchmarks Measure Capability, Not Impact

    Benchmark

    Fixed evaluation object

    Replayable input + baseline

    Defined metric, scope, and cost

    Production

    Incomplete, changing facts

    Constraints + conflicting qualities

    Consequences and ownership over time

24. 24. Form Improves Faster Than Architectural Coherence

25. 25. The Roadmap Begins with Knowledge Infrastructure

26. 26. Decisions Must Trace to Runtime and Back

27. 27. AI Analyzes. The Architect Owns the Trade-off

28. 28. Architecture Knowledge Is Not a Large Prompt

    Prompt

    One-run context snapshot

    Implicit version + freshness

    Weak provenance; no claim owner

    Living knowledge

    Schema + typed relations

    Versions + freshness policy

    Evidence + provenance + owner

29. 29. Walk One Change in Both Directions

30. 30. Start with a Workflow, Not Products

31. 31. What the Evidence Supports

    51 studies map the field, not ROI

    Narrow tasks show bounded value

    21 tools automate separate islands

    Benchmarks stop before production impact

    Start with living knowledge + human accountability

    Cheaper form raises the value of coherence, evidence, and ownership

32. 32. Architecture Memory Matters More Than Another Answer

    The longread, paper, and replication package are linked from the first slide

    Book Cube

    Continue the discussion on architecture, AI4SDLC, and engineering research in the channel.

    Alexander Polomodov, Technical Director & Fellow, T-Technologies

    @Book_Cube