--- title: "Dispatch: Field notes from Kensink Labs" description: "A lab dispatch from Kensink Labs. Engineering notes, half-finished diagrams, and the occasional rant about file formats. Twice a month, free, no tracking pixels." source: "https://www.kensink.com/dispatch/" canonical: "https://www.kensink.com/dispatch/" --- ★ Dispatch · v01.01 Lab open · 2 slots Q3 \+ 3 new case studies 047 engagements shipped # We build AI products on solid ground with direct calls to the model, boring infra, and senior people. Kensink Labs is a small lab of senior engineers shipping production AI for SaaS companies. We integrate against the model the same way we integrate against Postgres. No agent frameworks, no orchestration vendors, no migrations every six months. Eight weeks, then it’s yours. [Read this issue →](#dispatch-subscribe) [Browse case studies →](https://www.kensink.com/cases) Edition Vol. 01 · No. 01 Published 23 May 2026 Format Field notes · email + web Cadence Twice monthly \[FIG. 01 · ENGAGEMENT TIMELINE\] ## How an engagement is shaped, from discovery to handoff. Four overlapping phases over roughly sixteen weeks. Discovery feeds the brief; the brief feeds the build; the build feeds a runbook your team can still understand six months from now. DURATION · 16 WKS TYPICAL TEAM · 3 to 4 DRAWN · 22.05.2026 SHEET · A.01 **04** / 11 · INDEX OF WORK ## Recently filed. Six chapters, one lab. Three case studies you can read end-to-end, two essays we keep getting asked for, and one running archive of everything else. Pick the one closest to where you are now. CH. KIND TITLE SUMMARY LIFT READ [ 01. CASE STUDY Claims triage for a B2B insurer Direct-LLM pipeline · 12 weeks · Nuforce A direct-LLM triage layer over Nuforce's legacy claims pipeline. Reduced human review by seventy-three percent in twelve weeks. No agent framework, full ownership at handoff. 73% AUTO Read → ](https://www.kensink.com/cases/nuforce)[ 02. CASE STUDY An internal-research copilot, on Slack pgvector + citation-aware retrieval · Affidavit Mapp A retrieval copilot threaded into a fifty-person research team's daily workflow. Four hours saved per analyst per week within thirty days of rollout, measured at the team level rather than estimated. 4HRS / ANALYST Read → ](https://www.kensink.com/cases/affidavit-mapp)[ 03. CASE STUDY Onboarding that pays for itself by week one Adaptive prompts on the edge · AICoach An LLM-powered onboarding flow inside a consumer fitness app. Lifted week-one activation by eighteen points in the first eight weeks post-launch, with the eval suite the team now owns. +18PT ACTIVATION Read → ](https://www.kensink.com/cases/aicoach)[ 04. ESSAY Why most AI projects fail Six failure patterns we see in every audit Six failure patterns we see in every audit, ranked by how much money they cost, plus the cheap fixes that prevent them. 7MIN READ Read → ](https://www.kensink.com/blog/why-ai-projects-fail)[ 05. METHOD The Frontier Firm playbook A twelve-week field guide How a 200-person company restructures its org chart around AI agents without firing anyone, without losing a year of credibility, and without buying a transformation deck. 8MIN READ Read → ](https://www.kensink.com/blog/frontier-firm-playbook)[ 06. DISPATCH Every field note we've published Running archive · all entries Internal correspondence we found worth publishing. Mostly things we got wrong on the first pass and how we re-typeset them. 04ENTRIES Read → ](https://www.kensink.com/blog) 05 / 11 · LAB PRACTICE ## Three columns hold the lab up. We do three things, and we are honest about what they are. Every engagement starts in one column and usually ends in another. The scaffolding is how the work travels. P-01 / 03 · EMBEDDED ### Engineering teams, embedded. Two to four of our engineers join yours for an engagement of twelve to twenty weeks, with a published runbook and a working artefact at the end. We don't do retainers. Cycle 12 TO 20 WK Team 2 TO 4 PEOPLE Stack YOURS Output SHIPPED P-02 / 03 · CHAPTERS ### Product chapters, end-to-end. You bring a problem; we bring a chapter. Brief, build, ship and a printed companion runbook, by the date we agreed on, at the price we agreed on. Cycle 6 TO 8 WK Price FIXED Output APP + DOC Handoff WK 8 P-03 / 03 · DISPATCH ### Lab dispatches, made public. A field-notes practice on the side: short essays, the occasional open-source tool, the occasional public talk. The lab writes more than it speaks. Cadence 2× / MO Readers EARLY Format EMAIL · RSS Price FREE 06 / 11 · BY THE NUMBERS ## Four years. Two coasts. One stubborn lab. We publish these quarterly. The numbers don’t always flatter the lab, but they’re falsifiable. Footnoted, source-linked, updated 22 May 2026. ENGAGEMENTS · ALL-TIME 47 / 50 target across four years, two coasts, and one stubborn rule about what we don't take on. LAB HEADCOUNT 6 people senior engineers, designers, and one product lead who keeps everyone honest. REPEAT ENGAGEMENT 82 % of clients have shipped a follow-on chapter with us. Self-reported. RETAINERS HELD 0 ever every engagement ends on the date we agreed on. Then you decide. FIELD NOTES · 042 · 041 · 040 ## Recently filed, short essays from the lab. Engineering notes, half-finished diagrams, the occasional rant about file formats. New entries every other Wednesday in the Dispatch. NOTE · 042 20 MAY 2026 ### The file format is the API. A short brief on why our integrations team starts at the schema, never at the endpoint. Includes a diagram you can print at A4 and a fourteen-line Postgres migration. 7 min read [Read note →](https://www.kensink.com/blog/rag-vs-fine-tuning) NOTE · 041 06 MAY 2026 ### Costing a feature in pages, not points. Story points are an honest unit if you treat them like word counts. Here is how we write a feature like a chapter: outline, draft, edit, set, ship. 11 min read [Read note →](https://www.kensink.com/blog/llm-roi) NOTE · 040 23 APR 2026 ### On the printed README. Why every engagement ends with a physically printed runbook, and what changed about our work the day we started doing it. 5 min read [Read note →](https://www.kensink.com/blog/why-ai-projects-fail) DISPATCH · TWICE MONTHLY ## The lab dispatch. Worth keeping. SUBSCRIBE · FIELD NOTES, NO MARKETING Early readers · unsubscribe in 1 click · no tracking pixels NEXT ISSUE · 01 / 02 “Postgres is most of the system.” A working note on what we leave out of an AI stack, and why. 04 June 2026.