A healthy-looking Active players tile forced a sharper question about which sessions belonged in the metric.
ReportActive players looked real until we asked which sessions countedDEV ↗Articles / lines of reasoning
The writing returns to five problems.
Weekly AI Engineering Field Reports, written from systems I operate. Each line below opens with the argument that line has arrived at; the report that makes the case sits underneath it, and the body stays on DEV.
LINE 01
Measurement & Telemetry
3 reports · Codenames AI — co-primary
LINE 02
Authority & Stop Conditions
3 reports · Renovate governance; agent-native
Multi-slice agent plans need explicit authority handoffs and stop lines — not only implementation checklists.
ReportThe agent plan had every step except where to stopDEV ↗One report in this line — retiring a throwaway experiment repo — carries no system either. Three of fifteen reports sit this way; the line is paired, the row is not.
LINE 03
Evals & Judgment
4 reports · Editorial workflow — supporting
After fixing score-first critique, further reviewer gains came from separating kinds of reasoning rather than expanding the rubric.
ReportI fixed my AI reviewer. Then I kept solving the wrong problemDEV ↗LINE 04
Product Taste & Verification
3 reports · Codenames AI — co-primary
Coming back should restore what the product has accepted as true, not whatever was on screen — the persistence boundary is semantic, not architectural.
ReportThe board came back. The highlights lied.DEV ↗LINE 05
Agent Portability
2 reports · agent-native / editorial
Portability is a question of ownership and lifecycle, not a service boundary.
ReportI was solving agent portability at the wrong boundaryDEV ↗Neither report here documents a single system — they describe how agent setups move between projects. The line is paired, the rows are not.
Every report is published on DEV — full archive ↗