Field Notes - Dec 08, '25

Executive Signals

• Breadth is the new depth: first-submit across vendors exposes edges, hardens adapters
• Polling is the new webhook: enforce response SLAs, add events only when latency breaks
• Evidence over theater: traces and trendlines beat fragile live demos for trust
• Plans are living contracts: one-page weekly updates align teams through pivots
• Infra before polish: queues and state early, UI cosmetics after demo scope

CEO

Plans That Survive Contact

Static artifacts decay on first pivot. Keep a lightweight, continuously updated phase plan with written rationale so teams steer without ceremony. Empower engineering to propose changes, link plan items to tickets and acceptance criteria, and let the plan tighten as unknowns collapse rather than fossilize.

• Maintain a one-page phase plan; update weekly with owner and date
• Link milestones to checklists and acceptance tests; avoid static diagrams
• Require pivot rationales to align stakeholders when scope changes

Product

Biweekly WIP Demos With Guardrails

Lower the stakes by showing working slices every two weeks, but set expectations. Start each session with a 30-second frame on goal, scope, and allowed feedback. Route scope creep through one owner. Tag work as Requirement, Feature, or Optimization, and let only Requirements gate the phase. Time-box demo prep to ≤10% of the sprint.

• Open with goal, scope, allowed feedback; time-box prep to ≤10%
• Tag R/F/O; let only Requirements gate
• Channel scope requests through a single owner

Show the System, Not the Skin

For back-end automation, stakeholders need evidence, not theatrics. Replace fragile live runs with a tight narrative: screenshots of key states, an execution trace, a small adapter interface snippet proving extensibility, and a few hard metrics like throughput, error rate, and queue depth. Script asset capture so “demo assets” don’t consume a half day.

• Ship a one-pager “How it works” plus three metric trendlines
• Include a 10–15 line adapter contract snippet to prove extensibility
• Automate asset capture for consistent, repeatable demos

Engineering

Parameterize Compliance Cutoffs by Brand

Compliance windows vary by OEM; avoid hardcoding. Ship a single automation with a default cutoff, but make per-brand cutoffs and disposition logic configurable. Enforce cancel/allow decisions in adapters, capture a structured reference ID every time, and use free-text notes only when deliberately canceling outside the window. Provide a permitted test path per brand to submit, screenshot, and extract the reference ID.

• Build per-brand configs for cutoffs and statuses with sane defaults
• Enforce decisions in adapters; persist reason and reference IDs
• Provide brand-specific test flows to submit and capture evidence

Breadth-First Adapters Prove Architecture Faster

Going wide to “submit + reference ID” across OEMs surfaced integration edges early and forced a clean adapter interface. Doing one OEM end-to-end first risks ossifying design and slowing later additions. Measure time-to-first-submit for new adapters, and separately track the delta from “first” to “production-ready” to avoid hiding scope.

• Define and lint a strict adapter interface
• Optimize for first-submit across vendors before deepening
• Track first-submit-to-production delta as its own metric

Frontload Infra, Backload Polish and Set Demo Scope

Wire queues, orchestration, and state stores before features pile on; it’s cheaper and stabilizes later work even if the demo doesn’t need it. UI churn often comes from unclear demo expectations—publish “demo-ready” acceptance that excludes cosmetics, freeze noncritical UI two weeks before demo, and push polish to backlog. Treat DevOps checklists as parallel streams; block only what’s required to test scale.

• Stand up queues, orchestration, and state early
• Publish demo-ready acceptance; freeze noncritical UI two weeks out
• Run DevOps in parallel; block only scale-critical items

Delay Webhooks Until SLA Demands It

Minutes rarely matter. Poll at a sane cadence until a defined production response-time SLA is missed. Set response SLAs (e.g., 30–60 minutes) for compliance actions; if actuals exceed SLA for two consecutive weeks, then invest in webhooks. Align polling intervals to ≤50% of the SLA.

• Set and monitor response SLAs per action
• Keep polling until persistent SLA breaches justify events
• Tune polling cadence to half the SLA