AI-Native Codebases x BLR Kotlin

Transcript

1. April 4, 2026 • BLR Kotlin AI-Native Codebases Ragunath Jawahar Legacy Code Labs • legacycode.com
2. Agenda • Background • 4 Pillars of research • What is an AI-Native codebase? • Factors in fl uencing AI-Native development • Cognitive Debt • Case Study: Clarity
3. 4 Pillars of Research
4. 01 Mechanized Comprehension How large-scale software systems can be understood through tooling that extracts meaning, structure, and intent from code at scale. 02 Mechanized Veri fi cation How correctness and gaps in evolving software can be identi fi ed systematically and autonomously. 03 Mechanized Remediation How problems, once identi fi ed, can be addressed through targeted, tool-assisted intervention rather than expensive rewrites. 04 Directed Evolution How problems, once identi fi ed, can be addressed through targeted, tool-assisted intervention rather than expensive rewrites.
5. Outcomes
6. Match 3 Game: Jalebi Crush
7. Not AoE 2
8. 01 Mechanized Comprehension How large-scale software systems can be understood through tooling that extracts meaning, structure, and intent from code at scale. 02 Mechanized Veri fi cation How correctness and gaps in evolving software can be identi fi ed systematically and autonomously. 03 Mechanized Remediation How problems, once identi fi ed, can be addressed through targeted, tool-assisted intervention rather than expensive rewrites. 04 Directed Evolution How problems, once identi fi ed, can be addressed through targeted, tool-assisted intervention rather than expensive rewrites.
9. Implications • Improved throughput • Higher-level of abstraction • Programming languages • Frameworks • Platforms • Manage and work across multiple codebases • Behavior reviews over code reviews • Long running tasks and therefore higher spend on tokens • High AVOT (Autonomously Veri fi ed Output Tokens)
10. What is an AI-Native codebase?
11. An AI-Native codebase is one engineered primarily for contributions from coding agents and equally from human developers.
12. Factors In fl uencing Development 1. Developer skills 2. Coding agents 3. Codebase 4. Environment 5. Tooling 6. Feedback loops 7. Cognitive Debt
13. 1. Developer Skills • Level of abstraction • 🚩 Treating coding agents as junior developers • Shifting gears & experimentation • Tool building (dismantling identity)* • AI spends • Tool obsolescence
14. 2. Coding Agents • Experiment* • Spend on coding agents**
15. 3. Codebase • Stack • Guardrails • Forma tt ers • Linters (Default + Custom) • Git hooks • Structure • Replicate • Screaming architecture • Balancing act for humans and agents • A tt ributes • Ownership • AI Use • Prompts • Skills • Subagents
16. 4. Environment • Local agents • Background agents
17. 5. Tooling • Experiment • Removal over addition • Custom tooling
18. 6. Feedback Loops • Speed (long feedback loops ☠) • Granularity • Reliability • Cost 💰 💰 💰 • Who is the receiver? • Who’s closing the loop? • Humans • Agents • Environments • Tools
19. 7. Cognitive Debt
20. Development Patterns • Slicing • Horizontal • Vertical • Builds • Top-down (Outside-in) • Bo tt om-up (Inside-out) • ✨ Single-shot • ✨ Spike & stabilize discard
21. System Comprehension • PRD, TDD, ADR, RFC, etc., • Reviewing designs & user fl ows • Writing code • Reading existing documents • Speaking with colleagues • AI • Agentic code generation • Conversing with an agent • Reading code from an IDE • Reviewing PRs • ✨ Deliberate study time
22. You can't manage, fi x, or evolve a codebase you don't understand.
23. AI-Native Codebase: Clarity
24. 01 Mechanized Comprehension How large-scale software systems can be understood through tooling that extracts meaning, structure, and intent from code at scale. 02 Mechanized Veri fi cation How correctness and gaps in evolving software can be identi fi ed systematically and autonomously. 03 Mechanized Remediation How problems, once identi fi ed, can be addressed through targeted, tool-assisted intervention rather than expensive rewrites. 04 Directed Evolution How problems, once identi fi ed, can be addressed through targeted, tool-assisted intervention rather than expensive rewrites.
25. Clarity • Initially built as a review tool for AI-generated code • Clarity is a software design tool for developers and coding agents • Go (30K LOC) • Supports 15 programming languages • Recursively developed • Deterministic, no token usage
26. Installation Step 1: brew install LegacyCodeHQ/tap/clarity Step 2: clarity setup
27. Use Cases: Developers • Comprehension during development • Examine module and directory structures • Find paths between fi les (explore system edges) • Works best when you have one type per fi le
28. Use Cases: Coding Agents • Remediation • Identify • Missing abstractions • Incorrect abstraction • Find and fi x cyclic dependencies
29. Making Clarity AI-Native • ✅ Stack • ✅ Guardrails • 🟠 Contribution from agents and developers • ✅ Regression protection • ❌ Adding support for new languages
30. Approach 1. Structure 2. Behavior
31. 1. Structure
32. Structure: Before
33. Structure: After
34. 2. Behavior
35. Manual Work fl ow: Verify Language Support
36. Manual Work fl ow: Verify Language Support
37. SKILL.md (truncated) - - - name: verify-language-support description: End-to-end workflow for validating language support changes in the clarity dependency graph analyzer - - - Use this workflow to validate language support end-to-end in `clarity`. ## Prepare 1. Identify the target language module under `depgraph/<language>/`. 2. Identify affected command output in `cmd/languages/` when maturity level changes. 3. Confirm current local changes with `git status - - short`. ## Validate on a Real Repository (Always Interactive) 1. Pick a representative repo for the language. 2. Clone into `/tmp`. 3. Build a review queue before rendering graphs: - Use non-merge commits only. - Use commits with `5-30` changed files. - Prioritize commits that are mostly about the target language (based on file extensions/paths). - Each selected commit should include at least a few files in the target language (minimum 3 unless unavailable). - Default queue size is 10 commits unless the user requests a different count. 4. Show the queue to the user before starting graph renders.
38. Implications • Enable external contributions from agents and developers • Capture requirements from real-world codebases • Access global knowledge and expertise • Focus optimizing for human a tt ention, expertise, and judgement
39. References • Clarity (h tt ps://github.com/LegacyCodeHQ/clarity-cli) • Working E ff ectively with Legacy Code (Michael Feathers) • Refactoring (Martin Fowler) • Your Code as a Crime Scene (Adam Tornhill) • Software Design X-Rays (Adam Tornhill) • Glamorous Toolkit (h tt ps://gtoolkit.com/)
40. Questions? Ragunath Jawahar • h tt ps:legacycode.com @ragunathjawahar