The bug that comes back
Every team has a version of this story: a production incident gets fixed, everyone moves on, and eight months later the exact same class of bug happens again — different symptom, same root cause, fixed by someone who has no idea it already happened once.
The first fix wasn't wasted. What's missing is the second half of the work: writing down why it happened, so the next person doesn't have to re-derive it under pressure. That gap has a real, measurable cost — it just doesn't show up on the invoice for the first incident.
Where the cost actually lands
Re-investigation time. Someone re-runs the same diagnostic steps, checks the same logs, rules out the same false leads — work that was already done once and thrown away because it lived in one person's memory or a closed incident channel.
Re-breaking a fixed constraint. A workaround gets removed by someone who didn't know why it was there, because nothing recorded the reason. The bug it was preventing comes back, usually in production, usually discovered by a customer.
Bus factor becoming an actual outage. The person who remembers the reasoning goes on vacation, changes teams, or leaves. The knowledge doesn't get handed off — it just isn't there when the next incident needs it.
Repeated debate. A technical choice gets re-litigated every time someone new joins, because there's no record of what was already considered and rejected, and why. Every re-litigation costs a meeting, and sometimes reverses a decision that was correct the first time for reasons nobody in the room remembers.
Why this is worse than it looks on a roadmap
None of this shows up as a line item. It's distributed across a hundred small "wait, why does this work this way?" moments, each one a few minutes to a few hours, that never get attributed back to a missing wiki page from a year ago. That makes it easy to underinvest in — there's no single incident report that says "caused by undocumented decision from Q2."
The closest thing to a receipt is a root cause analysis that ends with "this happened because a prior workaround was removed without anyone knowing why it existed." That sentence appears in postmortems more often than most teams would like to admit.
What actually prevents it
Record the decision, not just the outcome. "We use polling instead of webhooks" is an outcome. "We use polling because the vendor doesn't support webhooks, and we tried and reverted a webhook-based approach in March" is a decision record — it survives someone re-proposing webhooks next year.
Attach it to the work, not a separate meeting. A decision written up during a scheduled "documentation day" months later has already lost most of its detail. The moment a task or an investigation closes is the cheapest time to capture it — see why documentation always falls behind for why timing matters more than intent.
Make the capture automatic, not aspirational. WKFGo's MCP integration asks — every time an AI agent closes a task — whether to write up what happened as a wiki page. It's the same discipline a good engineer applies manually, just asked consistently instead of only when someone remembers to.
A cheap test for your own team
Pick any workaround or unusual pattern in your codebase and ask someone on the team to explain why it's there. If the honest answer is "I'm not sure, it's probably load-bearing" — that's the cost showing up in real time, and it's worth a wiki page today, before it becomes tomorrow's incident.
FAQ
Isn't a decision log just more overhead? Only if it's a separate ceremony. Captured at the moment a task closes — which is what an "ask, don't assume" MCP nudge does — it costs one yes/no answer, not a meeting.
What's the minimum a decision record needs? The choice made, the alternative rejected, and the constraint that forced it. Three sentences is often enough.
Does this apply to AI-agent-written code too? Especially there — an agent's context window closes completely between sessions, so a decision it doesn't write down is gone as thoroughly as if a person forgot it.
Bottom line
The cost of an undocumented decision isn't the first incident — that one gets fixed either way. It's every hour spent re-discovering the same constraint after the person who found it is gone, distracted, or asleep.