Binding Vows with AWS: The Judgment AI Still Needs

AWS Community Day Taiwan 2026 https://awscmd.tw/en Serverless gives us real power, but only if we accept its limits. They cover execution time, resources, package size, concurrency, and API rates. In return we get managed services we did not have to build. Those limits also match what a good design should do anyway. Misread them, and the same design starts to cost us. We once said "fully serverless" and meant different systems. The word was shared; the constraints behind it were not. AI coding tools lowered the barrier to building on serverless. They still need the experience we never wrote down. This talk is that record.

Transcript

2. User Example AWS application API Gateway Lambda DynamoDB
3. User API Gateway Lambda What outcome makes this useful? Example AWS application DynamoDB
5. Structure AI builds Application People Domain + context Value People own the purpose and the context passed into the work
6. Structure Amazon Bedrock Code + IaC Lambda Structure carries our decisions Example AWS workflow DynamoDB
7. Old setting Rounding master From our team's records
8. Old setting Investigate Rounding master Who uses it? From our team's records
9. Old setting Investigate Omit Rounding master Who uses it? No operational need From our team's records
10. Normal process Record shipment completion From our team's records
11. Normal process Bypass route From our team's records Record shipment completion
12. Planned flow Work area Picking + movement Actual location From another team's record
13. Planned flow Work area Revise flow Picking + movement Actual location Discuss with team The space was narrower than assumed From another team's record
14. 02:00 after midnight Clock date Example dates Day D + 1
15. 02:00 after midnight Clock date Business date Example dates Day D + 1 Day D
16. Records AI Questions People code + history find patterns facts + gaps confirm meaning People decide value AI supports inquiry · people confirm meaning
17. Purpose Rule Example Correct business period Explicit reference date Missing date → error Owner Revisit Delegation Who decides? When work changes Agreed scope Keep it beside your README or PR checklist
18. Request Snapshot Expected Date + version reference date master data business result effective time Freeze the inputs before comparing A team proposal, under review
22. Business Port Adapter Lambda DynamoDB team owns rules GenRyu contract tested behavior example example Auth · dedupe · unknown commits GenRyu target design · in development
23. Codex: plan + review Claude: implement Human: merge / deploy GitHub Issue AI work Checks Review Draft PR TanRyu policy AWS MicroVM candidate SHA same candidate GitHub CI Tested on 3 GitHub Issues → 3 Draft PRs
24. Codex: plan + review Claude: implement Human: merge / deploy GitHub Issue AI work Checks Review Draft PR TanRyu policy AWS MicroVM candidate SHA same candidate GitHub CI Eligible check failure → bounded repair Tested on 3 GitHub Issues → 3 Draft PRs
25. Candidate Checks Review Evidence one SHA fixed checks fresh review saved evidence Verify the candidate you deliver A person makes the final call Draft PR
26. One request → one expected business outcome User API Gateway EventBridge SQS Lambda Expected effect · Deadline · Retry rule Agree on behavior before implementation DynamoDB
34. Accepted ID Worker Outcome store Safety Progress No duplicate or unauthorized effect Every ID meets the deadline Keep unfinished and unknown outcomes visible
35. Observe Decide Update Check Application signals + work rule owner context examples change Use outcomes to revisit rules Next step for GenRyu and TanRyu
36. 15-minute limit Agree Measure Record what finishes together? accepted limits predict behavior vow + reason One rule · One reason · One check Put the owner and revisit condition beside the README or PR checklist