Field Notes - Nov 03, '25

Executive Signals

AEO is the new SEO: build for overviews and carousels, not blogroll click farms
Update trains beat bursts: consistent releases move store algorithms and impressions
Retention is distribution: micro-fixes compound actives when installs stay flat
Attribution lies, accounting tells truth: reconcile WAU with net installs before celebrating
Plan, then build: agent briefs reduce drift and rework across repos

Weekly actives only grow as fast as net installs. If WAU is flat while installs trickle in, churn is matching acquisition. Big jumps with flat installs usually mean temporary algorithm boosts or measurement noise. Use ceiling math to choose the lever: increase qualified installs or reduce uninstalls, not hope for spikes.

Review 7/30/90-day installs, uninstalls, and WAU/MAU together each week
Set targets tied to the binding constraint: more installs or fewer uninstalls
Treat “algorithm lifts” as transient until proven durable

Causality Before Celebration

A 15–25% active-user uptick with flat installs often reflects auto‑update reactivation or external exposure. If a prior update didn’t spike, “update = spike” is a weak story. Open an incident, reconcile cohorts and referrers, and freeze changes until a single root cause is confirmed or falsified.

For >15% moves in 24 hours, start a 48‑hour “mystery spike” incident
Reconcile net installs, retention, store impressions, and referrers before claims
Require one root‑cause hypothesis; unfreeze only after confirmation

Marketing

Win the Extension Store Algorithm with Update Trains

Treat the extension store as an algorithmic surface. Repeated updates, refreshed creative, and precise keywords lift impressions and actives even when installs don’t spike. Monitor “Similar extensions” placement and competitor launch cycles to catch indirect traffic.

Ship every 3–4 weeks and change at least one public asset
Optimize for 5–8 exact search phrases; track impressions and CTR weekly
Watch Similar-placement and competitor updates for traffic shifts

Build for AI Overviews and Video Carousels

Short, focused how‑to videos earn carousel visibility and help win AI Overview citations. Recency plus authoritative depth drives inclusion. Pair these with refreshed cornerstone pages and living FAQs on home and product surfaces.

Publish 2–3 sub‑5‑minute how‑tos mapped to transactional queries
Stamp “Updated Month YYYY” and refresh cornerstone pages quarterly
Track 10–20 priority prompts; iterate schema, titles, and FAQs

Always Add One Open‑Ended Question to Forms

An optional “How can we improve?” field explains conversions better than checkboxes. Routed to CRM with source data, it exposes friction themes that convert directly into roadmap items.

Place it last; keep the prompt short; optional but visible
Pipe responses with campaign/source into CRM; review weekly for patterns
Turn repeated themes into backlog items with owners and dates

Sales

Give Event Leads Their Own Pipeline and 30/60/90 SLAs

At scale, event leads need a dedicated pipeline, clear ownership, and regional rules. Central trackers help, but conversion comes from disciplined hygiene and timed reviews. Separate reporting preserves focus and proper attribution.

Assign owner within 24 hours; publish regional disqualification rules
Enforce 30/60/90‑day follow‑ups with exit reasons at each gate
Report event MQLs separately to attribute wins accurately

Product

Retention Micro‑Fixes Beat Acquisition for Extensions

Small UX fixes created a step‑change in actives while installs/uninstalls held steady, signaling retention lift and store analytics lag. Make “active users” the north‑star and ship weekly micro‑fixes instead of waiting for big features.

Add an uninstall prompt; fix the top three reasons within a sprint
Hunt and resolve overlay/conflict bugs on popular sites and tools
Re‑check trends after 7–10 days against store impressions and potential migrations

Engineering

Plan–Then–Build Workflow for AI Coding Agents

Separate cognition from keystrokes. Have the agent draft a brief with objectives, interfaces, risks, and tests. Start a fresh thread with only that brief to build. The reset reduces tangents and yields cleaner diffs.

Require a 1–2 page implementation doc with acceptance criteria
Start a new chat: “Build exactly this” using the brief
Keep the brief in‑repo for reuse across agents and teammates

Make the Repo the Board

Make the codebase self‑narrating with markdown task files near each module. Agents and humans reference it as the source of truth, so progress survives chat resets and context limits.

Create /tasks.md per folder; link PRs; check off on merge
Prompt agents: “Read tasks.md; continue the next unchecked item”
Treat task changes as commits, not chat history

Rehydrate Agent Context from Commits, Not Memory

When threads drift or IDE updates nuke history, prime agents with code reality. Ask for a summary of the last 3–5 commits, infer current work, then proceed. This restores context and cuts hallucinations.

Prompt: “Summarize what changed in the last N commits; infer current work”
Then: “Implement next step per tasks.md and your summary”
Repeat the commit‑diff primer if the thread drifts

Use Higher‑Capability Models in Foreign Codebases

Exploration costs more than execution. In unfamiliar code, start with a stronger model to map the problem and draft the plan, then downshift once tests pass and anchors exist.

Use a high‑capability model for planning and the first PR
Measure by back‑and‑forth count and rework, not token price alone
Store planning briefs and tests in‑repo to enable handoffs