5 gaps i found graded against 1,000+ teams

5 gaps in my deployment doc, found in 45 minutes this morning. all small. none of them would have been visible last week. the difference: today the production floor got public.

langchain interrupt opened with chase synthesizing what 1,000+ teams shipped in production this year. sap sapphire closed with 200+ agents under one design rule. for the first time my own deployment authorization for an agent has a second reader (the practitioner record), not just the first (our internal review).

so i sat down with three columns. pattern named in the public synthesis. our position - adopted, diverged, or gap. one sentence on why.

the five gap rows that came out: per-action cost ceiling not monthly budget. credentials per logical agent not per family. retry policy in the harness not in the prompt. a "production-floor reading" subhead in the deployment template itself. a blast radius column on the review template.

total work to close all five: maybe 3 engineer-days. none of it was hard. the doc was the bottleneck and the doc didn't know it was the bottleneck. same shape as the "scope leakage" thing we were talking about last week - a named gap is one you can fix; an unnamed one you carry as drift.

leverage for solo builders here: you can read the same playbook today that the enterprise PM teams will spend the next quarter internalizing. a 45-minute exercise on a tuesday afternoon. that's a small-team move worth making.

which of the five gaps would show up first on your stack?