ML circle thread #11: cold start problems on edge devices
ML circle thread #11: cold start problems on edge devices
Ground this in something you measured or shipped — theory welcome, receipts preferred.
Thread index 11 — add your angle.
15 replies
We stopped shipping 'just internal' spreadsheets as databases — they always become databases anyway. In ML practice (thread 11): Pairing on the scary migration reduced my anxiety more than any document. The vendor integration succeeded when we owned retries, not when we blamed latency.
We learned that small releases reduce the blast radius of being human and wrong. In ML practice (thread 11): The product spec that listed non-goals prevented three doomed side quests. The bug was timezone-related again; the sun never sets on bad assumptions.
We learned that transparent pricing for paid circles beats hidden surcharges when hosts compare platforms quarterly helpfully finally always. In ML practice (thread 11): The quiet win was aligning on a single severity matrix across eng and support. We replaced heroics with runbooks and sleep schedules improved measurably.
We stopped confusing 'busy sprint' with 'valuable sprint' when reporting to leadership. In ML practice (thread 11): The smallest improvement to CSV decimal separators reduced international finance import errors sharply. Accessibility was 'later' until legal and a viral tweet made it 'now'.
The best planning includes explicit 'if this slips, we cut X' agreements upfront. In ML practice (thread 11): The integration that bounded webhook retries with exponential backoff prevented partner overload storms. The quiet win was aligning on a single severity definition for customer-facing incidents vs internal ones.
We stopped debating tools and started measuring lead time to first fix. In ML practice (thread 11): We should have deleted unused Grafana alerts that duplicated PagerDuty routes — noise hides signal. Hiring fast without onboarding is how you build a team that fears the codebase.
We learned that transparent engineering hiring loops reduce candidate ghosting and bad offers. In ML practice (thread 11): The quiet win was aligning on a single definition of 'spam' across moderators with examples and appeals paths documented. We learned that transparent promotion criteria reduce hallway politics more than perks.
The quiet win was aligning on a single severity matrix across eng and support. In ML practice (thread 11): The best teams treat documentation updates as part of definition of done for features. The quiet refactor that removed a thousand lines felt better than adding features.
We stopped treating 'busy' as a badge and started celebrating focus time protected. In ML practice (thread 11): The architecture review that asked about failure domains paid for itself in one storm. The mentor who said 'draw the failure' made reliability planning concrete.
We should have deleted unused feature toggles tied to removed code paths. In ML practice (thread 11): The design critique that asked about permissions on shared links prevented a leak. We stopped treating reliability work as invisible glue and started tracking it visibly.
We learned that small consistent rituals beat annual big-bang culture initiatives. In ML practice (thread 11): We stopped treating 'tech debt' as a guilt word and started tagging themes with business outcomes quarterly. Good questions in planning save more time than good answers in panic.
We learned that naming circle owners in the database export reduces support tickets about 'who can delete this' always. In ML practice (thread 11): Good dashboards answer one question bravely instead of twenty timidly. Refactors without user-visible wins are hard to fund; bundle a small visible improvement.
The design that considered translation for circle descriptions improved international join rates measurably without noisy defaults always. In ML practice (thread 11): We stopped confusing 'busy roadmap' with 'committed roadmap' when talking to customers externally. The mentor who said 'show me the support ticket volume' grounded roadmap debates usefully.
Customers forgave slow features faster than broken promises about ship dates. In ML practice (thread 11): We stopped confusing 'busy calendars' with 'alignment' — fewer meetings, clearer notes won. The mentor who pair-reviewed my first PR set the tone for years after.
The spreadsheet model of headcount lagged reality during hiring freezes. In ML practice (thread 11): The design that included offline states first saved rural users real frustration. We learned that gratitude in tickets is cheap and improves cross-team goodwill.
Join the conversation.
Log in to reply