The Problem
You've been using AI hard for six months. You added a code-review agent, then a ticket-triage agent, then something to draft your weekly digest, then a research assistant, then a thing that summarizes meeting transcripts. They all felt like wins when you set them up. You haven't looked at any of them since.
This week somebody asked you which of your agents was actually saving time and you couldn't answer. You think the code-review one still works. The digest one definitely went weird at some point — you've been quietly fixing its output for weeks. The research assistant you set up in March? You can't remember the last time you used it.
The discipline that protects a team from this drift is structural: sprint retros, postmortems, observability dashboards, a champion checking in on adoption. The SOLO mode has none of that scaffolding. The discipline that catches degradation on a team falls on a SOLO operator to invent for themselves — and almost nobody does. So agents quietly degrade, wins get slimmer, and you keep shipping without noticing that the AI part of your workflow is no longer pulling its weight.
This is also the trap for an individual contributor on a team. The team has its rituals; you still need yours. Team retros surface team-level patterns. They don't surface that your personal triage agent has been wasting your morning for two weeks.
The Core Insight
Rituals beat policies for individual practice. Small weekly disciplines compound; you can't policy your way into noticing that your agent stopped being good.
A team has structure to lean on. Sprint planning forces a pause every two weeks. Retros force a look-back. Postmortems force a write-up after incidents. Observability dashboards force the data to be visible. Even if no individual cares about discipline, the structure produces it as a side effect.
SOLO has no such structure. The only way to keep AI usage honest without a team is to invent personal versions of those rituals and run them on yourself. They have to be small enough to actually do, frequent enough to catch drift while it's small, and concrete enough that you know whether you did them this week or not. “I'll review my agents regularly” isn't a ritual; it's a wish. “Friday 3pm, 15 minutes, pick one random agent run and read the trace” is a ritual.
The four below are the ones that earn their keep. Skip them and you're flying blind on whether AI is actually adding value to your week. The lesson is the rhythm, not a fixed N-item canon — teams add a fifth or a sixth as their fleet shape demands.