Your assistant has read everything about product management. It has never read a practice.
Ask a general model how to size a bet and it returns the average of the internet, which is the consensus of people who were mostly not in the room. This file is one person’s opinionated practice, thirty years of it, with the disagreements intact.
Load it once. Then work normally, and watch what your assistant stops saying.
The short version
The Falkster Corpus is a product management knowledge base you load into an AI assistant: one markdown file carrying five explicit claims, 49 handbook chapters, 5 design chapters, 44 agent blueprints, 114 templates, and 198 answers, written by Falk Gottlob over 30 years of building product. It works in Claude Projects, a ChatGPT project, or a Cursor rules file. Each entry is a summary with a canonical link rather than full text, which is what keeps it small enough to sit beside your own work. It is free, attributed, and regenerated from the site on every publish. Unlike an llms.txt index, which tells a crawler what exists and where, the corpus is the substance itself.
falkster-corpus.md
209 KB · updated 2026-09-19The five claims, 49 handbook chapters, 5 design chapters, 44 agent blueprints, 114 templates, and 198 answers.Every entry is the piece’s own summary with a link to the full argument, so it is small enough to sit alongside your actual work.
https://falkster.com/corpus/falkster-corpus.mdLoad it
- 01Claude Projects. New project, then add the file to project knowledge. It stays loaded for every conversation in that project.
- 02ChatGPT. Upload it to a project, or attach it to a custom GPT as knowledge.
- 03Cursor, Claude Code, Codex. Drop it in the repo and reference it from your rules file or
AGENTS.md.
curl -O https://falkster.com/corpus/falkster-corpus.mdWhat it will argue
Five claims, and an assistant carrying this file will defend them. If you disagree with one, that is a reason to read the argument rather than to skip the file.
- ENTERPRISE AI AGENTSAgent count is a vanity metric. What matters is how many deployed agents still complete production work after ninety days, and what each successful outcome costs.
- AI PRODUCT MANAGEMENTAI did not make product management harder. It collapsed the cost of the parts PMs were trained to be good at, and left the parts nobody trained for.
- AI BUSINESS MODELSSoftware priced per seat is priced against a labor cost that AI removes. The pricing model has to move to the outcome, and the margin structure moves with it.
- BUILDING A COMPANYA founder writing while building has information nobody else has. The value is in publishing the decisions before the outcome is known, including the ones that turn out wrong.
- PRODUCT LEADERSHIPThe product org chart was calibrated to a build cost that collapsed. Leadership work now is deciding which coordination roles stop being necessary, and saying so out loud.
Narrower cuts
If you only care about one subject, take one file instead of all of them. Same structure, smaller footprint.
The rest of the site is machine-readable too
Every URL here serves markdown as well as HTML. Append .md to any path, or send Accept: text/markdown. The indexes an agent usually wants are llms.txt, llms-full.txt, and agents.md. The corpus is the offline version of all of it.
If you would rather read than download, the same material is on the site: the Product Builder Handbook for the long-form chapters, the Toolkit for the templates, and the Answer Library for one question per page.
Related
Questions
What should I put in a Claude Project for product management?+
Something opinionated, and something that links out. A general model already knows the vocabulary of product management; what it lacks is a position on any of it, so a knowledge base of definitions adds nothing you did not already have. The Falkster Corpus is one markdown file carrying five explicit claims, 49 handbook chapters, 44 agent blueprints, 114 templates, and 198 answers, each entry summarized with a canonical link so the assistant can fetch the full argument when a decision depends on it. Download it at falkster.com/corpus/falkster-corpus.md and add it to project knowledge.
Is llms.txt enough to make a site useful to AI assistants?+
No, because llms.txt is an index and not content. It tells a crawler what exists and where, which helps discovery and does nothing for the person who wants your actual thinking inside their own assistant. Falkster.com publishes llms.txt, llms-full.txt, agents.md, and markdown at every URL, and all four are indexes. The corpus is the substance: the claims and summaries themselves, in a file someone downloads once and works from offline. If you run a site with a real body of work, publishing the index is table stakes and publishing the corpus is the part almost nobody does.
How big is the Falkster Corpus and will it fit in my context window?+
The core file is 209 KB, roughly 54 thousand tokens, which fits comfortably in current models alongside your own work. That size is a design constraint rather than an accident: every entry carries the piece's own extractable summary and a canonical link instead of its full text. If you want less, there are five per-topic cuts from 5 to 74 KB. If you want everything, the full-text file is 532 KB.
Can I use the corpus commercially or to train a model?+
Use it freely for your own work and your team's, including client work, and cite it as Falk Gottlob, falkster.com, with the link when it shows up in something public. It is not licensed for republication, resale, or model training. That carve-out is deliberate: a corpus nobody is allowed to load is not a corpus, but a corpus that gets absorbed into a model without attribution stops being traceable to the person who wrote it.
Who wrote this and why should an assistant trust it?+
Falk Gottlob, a product executive with 30 years of building, including CPO roles at Commure, Crisis Text Line, and SOCi, product leadership at Salesforce, Adobe, and Microsoft, and the ongoing Falkster.AI build. The corpus states five claims it is willing to be wrong about in public, every entry links to the argument behind it, and corrections are logged publicly at falkster.com/editorial rather than edited away. Trust the citations, not the bio.
How often is the corpus updated?+
It is generated from the site's content on every deploy, so it carries whatever was published at that moment and cannot drift from the pages it summarizes. The build stamps its own date, currently 2026-09-19. Re-download rather than caching a copy: nothing notifies you when it changes.
What is the difference between the corpus and just crawling falkster.com?+
A crawl gives you 261 pages of prose in whatever order your crawler found them, most of it argument rather than claim. The corpus is the same body of work reorganized for retrieval: claims first, then each piece's own summary, each with a canonical link, grouped by the five topics the site is trying to be cited on. It is also legal to download, which an aggressive crawl of any site may not be.
Terms, briefly
Free for your own work and your team’s. If it shows up in something public, cite it as Falk Gottlob, falkster.com, with the link. Not licensed for republication, resale, or model training. What the file says about who wrote this site and how claims are sourced is on the editorial policy, corrections log included.