slckHow to Turn a Slack Conversation Into a Ticket
Turn a Slack conversation into a ticket with repository context, acceptance criteria, ownership, and traceability that keep engineering work moving daily.
Practical writing on running, supervising and shipping work with AI coding agents.
slckTurn a Slack conversation into a ticket with repository context, acceptance criteria, ownership, and traceability that keep engineering work moving daily.
workflowTurn an Azure DevOps work item into an AI coding agent session with the context, approvals, checks, and traceability needed for production delivery now.
workflow-continuityA Linear issue is rarely a complete implementation plan. How to turn one into a scoped, supervised agent session — context assembled, repository resolved, and every diff reviewed while it is still cheap to change.
A ticket that says "Add SSO support" is a starting signal, not an agent-ready task. Turning a Jira item into a bounded, reviewable agent session — with a human in control at every stage.
code-reviewRemote approval is not shrinking a desktop code review into a phone. It is a decision packet: enough context to make a defensible call, enough proof to verify it.
Our dashboard said we had eight paying customers. Four of them were us. What we found auditing our own analytics, and the honest version of the numbers.
A 0.91% trial conversion is not a copy problem in disguise. So instead of rewriting the paywall, we emailed the other 109 and asked one question.
The expensive part is not the code an agent writes. It is the context it re-reads — so cost grows with turns, not with lines changed.
An instant failure means the request was rejected before anything was built — and the HTTP status will probably not tell you why. A five-minute triage.
It took four deploys to server-render one page. Three of the four failures returned 200 OK — a successful status is not evidence of a working page.
engineeringTwo tests, both passing, both well named, both asserting the wrong behaviour. Neither was found by reading them — and coverage numbers cannot see either.
57.8% of users who chose a deploy target never finished. There were no error reports — the error dashboard was clean because of the bug, not despite it.
An agent can produce a diff in ten minutes. The hard problem starts at minute 45, still tracing a bug while its owner is in transit. Monitoring is a control loop.
Async agent work is not unattended delivery. It plans for the developer being away: structured context up front, visible checkpoints, and approvals backed by proof.
An agent can produce a useful diff in 20 minutes, then sit idle for six hours. The constraint is not model capability — it is continuity across people and time.
agent-supervisionRemote AI supervision is the operating layer between a stranded agent session and an unreviewed merge: see the work, keep its context, and decide from anywhere.
A coding agent does not stop being your responsibility because you closed your laptop. What remote supervision really requires is context and control at the moment a decision is needed.