The reading list
Essays, tech notes, and the in-between.
- Product·3 min read
Connecting the Tools: Tracker, Docs, and Chat Without Drowning in Integrations
The repo holds context and the tracker holds work — but your AI assistant needs to reach both in the same session, plus the docs where specs get collaborated on and the chat where decisions actually happen. This post wires it: MCP connections to Jira/Linear, Notion/Confluence, and Slack/Teams, the read/write rules that prevent a sync monster, and the permission boundaries you set before connecting anything.
- Product·3 min read
The Signal Pipeline: Tickets, Calls, and CRM into One Weekly Digest
Customer evidence is the most valuable input a product team has and the worst-managed: tickets in the help desk, transcripts in the call recorder, lost reasons in the CRM, all unread past week one. This post builds the pipeline — a signal card format, a per-source extraction pass, and a one-hour weekly ritual that turns raw exhaust into evidence your stories can cite.
- Product·3 min read
Designing the Context Repo: a Structure Agents and Humans Can Navigate
The difference between a context repo that compounds and one that rots is decided in the first week, by structure: small files that answer one question each, frontmatter that makes them queryable, decision records with dates, and an AGENTS.md that teaches AI tools how to read the whole thing. The complete layout, with copy-paste starter files.
- Product·3 min read
The Product Context Repo: Why Your Product Knowledge Belongs in Git
Every AI tool you point at your backlog produces confident, generic garbage — not because the model is weak, but because it has never met your product. The fix isn't a better prompt. It's a repo: a plain-text, version-controlled home for everything your product knows about itself, structured so both humans and agents can read it. Part one of a six-part build-along.
- Product·3 min read
Cut Too Deep: Klarna, Duolingo, and the Cost of Borrowed Conviction
In February 2024, Klarna announced its AI assistant was doing the work of 700 agents; fifteen months later its CEO told Bloomberg they'd 'cut too deep.' A year after that, Duolingo put AI usage into performance reviews, watched its most loyal users revolt, and quietly took it back out. Neither company was wrong about AI. Both were wrong about what their own principles required — and the distance between those two mistakes is where this whole series has been heading.
- Product·3 min read
The MVP Canon, Checked Against the Record: Buffer, Zappos, Dropbox, ChatGPT
Every product talk retells the same three MVP stories — Dropbox's video, Zappos's shoe photos, Buffer's landing page — usually with numbers nobody in the room has ever checked. So I checked. Buffer holds up on contemporaneous evidence; Zappos is real but retroactively labeled; Dropbox's famous waitlist figure is founder-self-reported and attached to the wrong video half the time; ChatGPT's 100M is an analyst estimate wearing a fact's clothing. The discipline survives the embellishment — but a PM quoting unverified numbers in a strategy deck is committing the exact sin the MVP exists to prevent.
- Product·3 min read
Sonos Shipped the Date, Not the App
In May 2024, Sonos replaced its app with a ground-up rewrite that shipped without alarms, sleep timers, and accessibility features, reportedly to hit the launch window for its first headphones. The invoice arrived on a schedule: a public CEO apology in July, around 100 layoffs and a $20–30 million remediation estimate in August, a 16% revenue drop in the SEC-filed fourth quarter, and the CEO's exit on January 13, 2025. The freshest, best-documented case of an organization shipping the deadline instead of the product.
- Product·3 min read
The Roadmap Is a Promise: Osborne, Star Citizen, and the $39,900 Cybertruck
In 1983, Osborne Computer pre-announced its next machine and was bankrupt within months — a story so scary it still governs release comms forty years later, even though it's only half true. In 2022, Star Citizen's developer deleted everything beyond the next quarter from its roadmap because 'players interpret anything on the release view as a promise.' And the Cybertruck that was announced at $39,900 arrived four years later at $60,990. Three eras, one mechanism: audiences do not perceive confidence intervals, so the only roadmap formats that survive are the ones that encode uncertainty structurally.
- Product·3 min read
Sales Sold It: The Recovery Playbook for Unplanned Commitments
The contract is signed, the feature is due in eight weeks, and nobody on the product side has ever scoped it. The two obvious reflexes — bend the quarter or refuse on principle — both make the next one worse. Here's the recovery playbook: establish what was actually promised, get in front of the customer fast, price the promise for the people who own the tradeoff, and turn the whole incident into roadmap signal instead of roadmap debt.
- Product·3 min read
Runway, Burn, and the Default-Alive Question
Paul Graham found that most founders can't answer the only financial question that matters: at your current growth and burn, do you reach profitability before the money runs out? Inside corporations it's worse — internal ventures have fake runway, a budget line instead of a bank balance, and no clock at all. This closing piece of the series is about the honest clock: burn mechanics, the burn multiple, why corporate ventures need artificial scarcity, and which levers to pull, in which order, when the runway shrinks.