We learned that humour helps retrospectives if it does not punch down at roles. In AI startups (thread 3): We learned that naming on-call secondary is as important as naming primary. The architecture review that asked about child safety workflows for public circles changed moderation staffing plans before launch measurably helpfully quarterly always.

We learned that writing 'success metrics' in RFCs prevents post-launch arguments about impact. In AI startups (thread 3): We should have load-tested the auth path before Black Friday, not after. We should have deleted unused Grafana alerts that duplicated PagerDuty routes — noise hides signal.

The quiet win was aligning on a single definition of 'done' across design and eng. In AI startups (thread 3): We should have invested in offline-friendly read modes before pitching global teams with unreliable connectivity honestly quarterly. We learned that kindness plus accountability is the combo that actually ships quality.

We learned that transparent incident timelines reduce conspiracy theories internally too. In AI startups (thread 3): The smallest improvement to thread notification batching reduced email fatigue for active circles quietly. We learned that small consistent rituals beat annual big-bang culture initiatives.

The quiet win was aligning on a single definition of 'active contributor' across circles and profiles finally. In AI startups (thread 3): We learned that transparent salary correction processes reduce quiet quitting risk measurably. We learned that writing 'communication plan' in launch checklists reduces stakeholder surprise always.

The design that considered colour contrast early passed audits without emergency heroics. In AI startups (thread 3): We learned that naming incidents consistently helps analytics later more than clever titles. We stopped shipping 'just internal' APIs without authentication because internal becomes external eventually.

We should have asked legal earlier about data residency — assumptions were expensive. In AI startups (thread 3): The integration that bounded LLM token budgets per request prevented runaway costs under abuse quietly. The smallest UX copy change cut support tickets more than a new API field.

The vendor demo lied by omission; our staging environment told the truth. In AI startups (thread 3): The architecture review that asked about backup restores caught a real gap. We stopped confusing 'agile' with 'no planning' when stakeholders were nervous.

We should have deleted dead feature code before the security review found secrets in it. In AI startups (thread 3): Readable logs beat clever logs when you are tired at three a.m. We learned that customers notice when performance improvements ship without fanfare — they feel it.

The flaky test order dependence taught us to randomise test order in CI finally. In AI startups (thread 3): We learned that transparent roadmap voting inside trusted circles produces better priorities than executive-only stacks often. We learned that customers trust circles more when moderators publish clear norms and enforce them kindly consistently.

We learned that repeating the same retro topics means we are not learning. In AI startups (thread 3): The product looked done at eighty percent and was actually forty percent of the work. We stopped treating accessibility as a polish pass and caught issues earlier.

We learned that writing 'non-goals' in RFCs prevents zombie scope resurrection. In AI startups (thread 3): The right default in config beats a thousand-page admin guide nobody reads. The smallest UX copy change cut support tickets more than a new API field.

The architecture spike that listed compliance constraints early saved redesign pain later. In AI startups (thread 3): The quiet win was aligning on a single definition of 'active user' across teams finally. The mentor who said 'write the customer apology draft before launch' improved incident comms.

The product looked done at eighty percent and was actually forty percent of the work. In AI startups (thread 3): The right default in config beats a thousand-page admin guide nobody reads. We stopped shipping 'temporary' IP forwarding rules that became permanent attack surface quietly.

We should have deleted unused CI secrets after rotating tokens — scanners found them anyway. In AI startups (thread 3): We learned that transparent promotion timelines reduce anxiety more than surprise bonuses. We learned that humour about legacy migrations is therapeutic if it ends with a concrete lesson learned.