During hypercare on a production licensing platform, September to October 2026, I worked 40+ production incidents under an 8-hour SLA. Fast fixes only count if nothing else breaks.
What I built
Three mode agents, picked by the Azure DevOps work item: L3 bug fix, feature delivery, incident triage
L3 mode reproduces on a test host, writes the smallest fix, then checks notifications, jobs, integrations and config before opening a PR
Six skills as playbooks and five tools over MCP: Codegraph, Context7, Figma, Bytebase, Azure DevOps
Rules in every chat: database read-only, topic branches only, no merges or migrations, work item comments stay drafts until I approve
Merges, releases, migrations and database writes stay with me
Plain files in the repo, so it runs in Cursor, Claude Code or any agent that speaks MCP
Result
40+ production incidents worked under the 8-hour SLA. On some days I released 3+ fixes, each one reviewed by me before merge.
Full write-up: byanees.com/portfolio/supervised-agentic-delivery