Field Notes are fast, from-the-trenches observations. Time-bound and may age poorly. Summarized from my real notes by . Optimized for utility. Not investment or legal advice.
░░░░░░░▄█▄▄▄█▄ ▄▀░░░░▄▌─▄─▄─▐▄░░░░▀▄ █▄▄█░░▀▌─▀─▀─▐▀░░█▄▄█ ░▐▌░░░░▀▀███▀▀░░░░▐▌ ████░▄█████████▄░████ ======================= Field Note Clanker =======================
⏺ Agent start
│
├── 4 data sources
└── Total 36.3k words
⏺ Spawning 4 Sub-Agents
│
├── GPT-5: Summarize → Web Search Hydrate
├── GPT-5-mini: Score (Originality, Relevance)
└── Return Good Notes
⏺ Field Note Agent
│
├── Sorted to 5 of 7 sections
├── Extracting 5 key signals
└── Posting Approval
⏺ Publishing
┌────────────────────────────────────────┐ │ Warning: Field notes are recursively │ │ summarized by agents. These likely age │ │ poorly. Exercise caution when reading. │ └────────────────────────────────────────┘
Viral loops amplify momentum; they do not manufacture it. Greenlight network-effect work only after the core experience shows pull and forgiveness. Read cohorts for shark fins and fix the value delivery before adding growth mechanics. Use NPS as the integrity check because it is harder to game and tracks forgiveness when things break.
Gate loops until NPS ≈ 50 and D30 retention clears an internal bar
Triage swelling-then-sliding cohorts before any virality spend
Optimize for more 9–10s and fewer 0–6s, not just the average
Hiring is cheaper than firing. Run a process designed to disqualify quickly against must-have competencies, not to accumulate soft yeses. Force real work to expose execution, strip away charisma bias, and prevent post-hire surprises.
Give every panelist veto power on three non-negotiable competencies
Require a paid work sample or shadow day for GTM roles
Ban gut overrides unless new evidence changes the signal
Discovery is shifting into assistants that quote and summarize in‑app. Content must be answer-grade, machine-readable, and grounded in first‑party proof. Ungate education to earn citation, then capture on higher‑intent second CTAs like tools and benchmarks.
Ship answer‑first pages with 100–200 word abstracts, JSON‑LD, and full transcripts
Publish living comparison and integration pages anchored in first‑party data
Allow AI crawlers intentionally with agent-specific robots rules and log reviews
You are too close to your own product. Borrow customer language to reduce distance and increase clarity. Treat intros earned as the objective measure of resonance, not internal taste.
Ask 30 customers for the three bullets they would use to intro you
Test those phrases as hero copy and CTAs; keep what earns intros and demos
Refresh quarterly as customer language drifts faster than your site
Creation is expensive; distribution compounds. A single live session can fuel a month of consistent, channel‑native publishing that keeps you present in feeds and answers.
From one 45–60 minute session: transcript → 4 posts → 12–15 LinkedIn posts → short clips
Operate a 3:1 repurpose‑to‑create ratio and schedule daily outbound
Pair with evergreen how‑tos and templates to capture latent demand
Cold email decay is real. Keep email, but only when it reflects intent, timing, and genuine personalization sourced from your CRM and product telemetry.
Pause sequences if reply rate <1% or spam complaints >0.1%
Trigger from behavior and milestones, not a calendar
Offer multiple conversion paths per touch to match buying stage
Misconfigured CRM rules silently demote real MQLs, starving ad platforms of feedback and breaking optimization loops. Sales can keep strict SQL gates while Marketing preserves the signals systems need.
Appoint a single CRM owner; run a monthly scoring audit and change control
Separate MQL from SAL/SQL; do not let firmographic gates demote intent
Pipe MQL and conversion events back to platforms within 24 hours
Small budgets get taxed by fixed fees, low ACV stalls payback, and LinkedIn punishes cold. Nail unit economics before scale and assign channels to their strengths.
Keep agency fees ≤20% of media with performance holdbacks and a 90‑day kill
Require initial contracts alone to clear ≤6‑month payback on gross margin
Use LinkedIn for retargeting or ABM; cap until CPL holds
You will not code your way to signal. Run a tight loop that onboards, earns intros, and compounds warm pipeline. Ship only fixes while you learn how outcomes resonate and which integrations are deal‑critical.
15‑minute onboarding → ask for three intros → book next demos; track intros per call
Sprint to ten successful onboardings; no new features, only bug fixes
Pre‑commit week‑buildable integrations when they are deal‑critical
Two link models serve different GTM jobs. Replay links validate freshness for self‑serve and engineering buyers. Snapshot links fix the moment for sales handoffs and controlled follow‑ups.
Use replay links for self‑serve validation; always current
Use snapshot links for handoffs; expire after N views or hours
Gate rescans per visitor to create demo‑worthy scarcity
Early funnels die on cognitive hurdles, not code. Delete abstractions like “workspaces” and “projects,” infer sane defaults, auto‑connect obvious integrations, and land new users on a screen that proves value immediately. Lowering effort beats trying to manufacture motivation; move work from the user into the product.
Target 80% of new users reaching first value in under 120 seconds
Replace multi‑step wizards with one click and prefilled actions
Auto‑connect common data sources and show live data immediately
Marketing visitors bounce fast. Treat scans like LLM responses: visible progress in under a second, incremental insights every couple seconds, and reactive UI. Cap depth, emit findings page‑by‑page, and attach duplicate users to the same job.
Time‑to‑first‑signal <1s; update every 1–2s with coarse metrics first
BFS up to ~10 pages; broadcast shared job state across subscribers
Let the score jitter as new pages land to show progress
Anything callable from a public site is effectively public. Assume attackers can replay and farm rescans. Put defense at the edge and keep secrets off the client.
Bot checks before API calls; backend per‑IP/per‑origin rate limits and 429s
Use scoped bearer tokens and WAF rules; rotate often and ignore Referer
Collapse duplicate rescans into a single backend job
Two caches create bugs. Keep canonical caching in the backend keyed by URL and location. If the website needs shareable reports, persist dedicated snapshots with their own lifecycle.
Ship the backend as a thin data plane first. Return raw lists and minimal metadata; compute gauges and top risks client‑side so heuristics can evolve without redeploys.
Define a stable v1 schema of four arrays plus minimal metadata
Keep derived scores in the frontend to change rules quickly
Add detailed mappings like asset‑to‑country in a later iteration
Misalignment hides in tiny decisions. Publish a single source‑of‑truth spec per sprint and freeze it. Changes queue to the next cut and ship with a changelog and reference client.