AI and APIs: Ensuring Platform Migration Reliability with AI
summary
Migration aims to prove the new API matches legacy semantics for a half‑billion‑euro business by automatically comparing huge, differently structured JSON responses. Prototype uses VS Code + Clin agent, MCP model context protocol with JSON Tools to let the LLM query JSON like a database, avoiding huge context windows. Full-LM checks are slow (~minutes, ~$1 per run) and flaky, so the final approach generates code offline (deterministic, cheap to run) and supports MCP JSON querying for large payloads and progressive 5% phased rollout.
AI-written summary · about
chapters (20)
- 0:00AI and platform migration intro
- 1:12Managerial pressure on AI
- 2:59Migration challenge
- 5:02End-to-end testing importance
- 6:47AI for API translation
- 7:53Proof of concept to project
- 8:35Breaking knowledge silos
- 9:40Too many technical solutions
- 10:21JSON context with MCP
- 11:45Handling discrepancies between legacy and new APIs
- 12:49AI approach limitations
- 13:52AI-generated test code
- 15:15AI projects vs traditional
- 16:21Embracing AI uncertainty
- 17:26Cultural impact of AI
- 18:28Knowledge transfer with teams
- 19:10Key lessons from AI project
- 21:12Q&A on AI testing
- 23:58Proving API migration reliability
- 25:05Gradual migration strategies
related talks
26mCan AI Build Your Next API Integration
43mAI-Generated API End-to-End Tests: A Legacy Rewrite Case Study
30mAI First Legacy Migration
28mAI Migration at SCALE: Make Complex Code Tasks 10X Faster (The Blueprint)
43mAI accelerated legacy modernisation
27mAI as a Service Scaling AI with Ease and Innovation
14mAI-accelerated migrations at scale
29mAPI Integration augmented by GenAI