Reposition as a knowledge base (v2.0.0): operator lessons in, skills out#2
Conversation
Repositions EdTech Founder Stack from an interactive-skills toolkit into a curated, AI-friendly markdown knowledge base for edtech founders. - Add data/operator-lessons.md: 71 lessons distilled and attributed from the free public archive of Lenny's Podcast and Lenny's Newsletter, mapped to selling into schools, universities, and L&D. - Remove the 11 interactive skills and the setup script. - Rewrite README, ARCHITECTURE, CLAUDE.md, CONTRIBUTING; update templates, CODEOWNERS, and the social preview; bump to 2.0.0. Note: .github/workflows/validate.yml still references the removed skills and is left unchanged here, because updating a workflow file needs the gh token's `workflow` scope. Grant it (gh auth refresh -s workflow) to land the CI cleanup. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7df5c4b515
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| EdTech Founder Stack is now a curated, AI-friendly knowledge base for edtech founders — markdown you point your AI tools at or read directly — rather than a set of interactive skills. The knowledge in `data/` is the product. | ||
|
|
||
| - **New `data/operator-lessons.md`** — 71 field lessons on validation, product, GTM, sales, pricing, pilots, fundraising, and team, distilled and attributed from the free public archive of Lenny's Podcast and Lenny's Newsletter and mapped to selling into schools, universities, and L&D. Paraphrased under the source's personal/non-commercial terms. | ||
| - **Removed the interactive skills** — the `skills/` directory and the `setup` script are gone; the `data/` knowledge base is the whole product. CI now validates the data and research files only. |
There was a problem hiding this comment.
This release note says CI now validates only data/research, but the checked workflow still runs .github/workflows/validate.yml lines 15-33 against skills/*/SKILL.md. Because this commit deletes every skills/ file, the glob is left literal and the frontmatter step exits with 2 frontmatter errors found, so every PR/push using this workflow is red until the workflow is changed or the skills directory is retained.
Useful? React with 👍 / 👎.
|
|
||
| MIT. Use these skills however you want. | ||
|
|
||
| MIT — use the knowledge however you want. The operator playbooks in [`data/operator-lessons.md`](data/operator-lessons.md) are paraphrased and attributed from the free public dataset of Lenny's Podcast and Lenny's Newsletter, under its personal/non-commercial terms; see the note at the bottom of that file. |
There was a problem hiding this comment.
For commercial edtech founders, this line grants “MIT — use the knowledge however you want” while the same sentence says the new operator playbooks are based on material used under personal/non-commercial terms. Since data/operator-lessons.md is now part of the shipped knowledge base, users cannot tell whether they may commercially reuse that file under MIT; either exclude it from the MIT grant or remove/replace the restricted-source material.
Useful? React with 👍 / 👎.
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request transitions the EdTech Founder Stack from an interactive set of terminal-based skills to a curated, AI-friendly markdown knowledge base. All interactive skills under the skills/ directory and the setup script have been removed, and the documentation has been updated to reflect this new structure. Additionally, a new file data/operator-lessons.md has been introduced, containing 71 distilled operator lessons. The reviewer suggested clarifying a link description in the new operator lessons file to better manage reader expectations.
|
|
||
| **Chase the buyer who pulls it out of your hands.** Hold a strong thesis about how the world is changing, but stay loose on the exact wedge and go after the customer who is shockingly easy to sell. If selling the next customer is a grind, the business won't scale — so trust the eager buyer over the district you spend nine months persuading that your pedagogically "correct" product is right. — Brendan Foody, *Lenny's Podcast* | ||
|
|
||
| **You can't research your way to a zero-to-one product.** People answer from what they've seen — ask about a touchscreen phone and they describe a better keyboard. Feedback loops are right for improving an existing category and actively mislead on category-defining bets. For an incremental tool, lean on educator feedback; for a new bet, expect teachers to ask for the familiar workflow and validate by putting a working prototype in front of them. — Caitlin Kalinowski, [*Lenny's Podcast*](https://www.lennysnewsletter.com/p/why-were-at-the-beginning-of-the) |
There was a problem hiding this comment.
For improved clarity, when linking to a newsletter article that is a write-up of a podcast episode, consider making the link text more explicit. This helps manage reader expectations about the link's destination.
For instance, you could adjust the link text to clarify it's a write-up or transcript. This pattern could be applied to other similar links in this file.
| **You can't research your way to a zero-to-one product.** People answer from what they've seen — ask about a touchscreen phone and they describe a better keyboard. Feedback loops are right for improving an existing category and actively mislead on category-defining bets. For an incremental tool, lean on educator feedback; for a new bet, expect teachers to ask for the familiar workflow and validate by putting a working prototype in front of them. — Caitlin Kalinowski, [*Lenny's Podcast*](https://www.lennysnewsletter.com/p/why-were-at-the-beginning-of-the) | |
| **You can't research your way to a zero-to-one product.** People answer from what they've seen — ask about a touchscreen phone and they describe a better keyboard. Feedback loops are right for improving an existing category and actively mislead on category-defining bets. For an incremental tool, lean on educator feedback; for a new bet, expect teachers to ask for the familiar workflow and validate by putting a working prototype in front of them. — Caitlin Kalinowski, [*Lenny's Podcast* (write-up)](https://www.lennysnewsletter.com/p/why-were-at-the-beginning-of-the) |
Addresses PR review feedback: - License (Codex P2): the MIT grant no longer blankets data/operator-lessons.md. README and LICENSE now state that file is personal/non-commercial only, since it is paraphrased from Lenny's Podcast/Newsletter under the source's terms. - Links (Gemini P3): the operator-lessons header now notes that lesson links point to the episode/post page on Lenny's site. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Drops the SKILL.md frontmatter and CLAUDE.md routing checks now that skills/ is gone; keeps the data-file footer and research-table validation. Resolves the red CI on this branch (Codex P1). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
All three review findings are addressed and CI is green:
@codex review |
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request transitions the EdTech Founder Stack from an interactive set of AI skills to a curated, AI-friendly knowledge base of markdown files, removing the interactive skills directory and setup script while introducing a new operator playbooks file (data/operator-lessons.md). The review feedback focuses on improving documentation maintainability and text clarity. Specifically, the reviewer recommends replacing hardcoded counts of papers and lessons with generic phrasing to prevent future documentation drift, ensuring consistency in sorting instructions, and addressing minor grammatical errors, typos, and phrasing issues within the newly added operator lessons.
| ### Learning science research — `data/research/` | ||
|
|
||
| ## Skills | ||
| 376 peer-reviewed papers across 19 topics: spaced repetition, cognitive load, formative assessment, adaptive learning, worked examples, the learning-styles myth, and more. Each paper carries its takeaway, study type, year, citation count, and DOI. When you tell a buyer or investor that something works, cite the paper, not a vibe. Index: [`data/research/README.md`](data/research/README.md). |
There was a problem hiding this comment.
In version 1.4.0, the project established a "Count-agnostic documentation" policy to prevent documentation from drifting as the corpus grows. However, hardcoded counts like 376 peer-reviewed papers (here and in ARCHITECTURE.md / CLAUDE.md / docs/social-preview.html) have been reintroduced.
Consider referencing the index or using a more generic description (e.g., "hundreds of peer-reviewed papers") to maintain count-agnostic documentation and avoid future drift when new papers are added.
| ### Operator playbooks — `data/operator-lessons.md` | ||
|
|
||
| **`/accessibility-check`** — Don't treat accessibility as an afterthought. Check against WCAG 2.1 AA, Section 508, and UDL. Get a prioritized checklist, VPAT guidance, and an action plan. | ||
| 71 field lessons on validation, product, go-to-market, sales, pricing, pilots, fundraising, and team — distilled from operators and investors on Lenny's Podcast and Lenny's Newsletter, and mapped to selling into schools, universities, and L&D. Practitioner experience, attributed and paraphrased; the research corpus is the evidence layer. |
There was a problem hiding this comment.
Similarly to the research papers count, the hardcoded count of 71 field lessons (here and in ARCHITECTURE.md / CLAUDE.md / docs/social-preview.html) will require manual updates across multiple files whenever new lessons are added.
Consider using a more generic phrasing like "dozens of field lessons" or referencing the file directly to keep the documentation maintainable and aligned with the count-agnostic goals.
| 5. **Smart routing.** Don't just list all the other skills at the end. Recommend the *one* skill they actually need next based on their scores or current stage. | ||
| 1. **Cite the source.** Every factual claim should trace to something checkable — a named regulation with a date, a paper with a DOI, a named operator. No guessing. | ||
| 2. **Be opinionated.** We don't do "on the one hand, on the other hand." If you add guidance, give it a point of view backed by real experience. | ||
| 3. **Keep the format.** Match the structure of the file you're editing. For regulatory and market files, keep the "last updated" date at the bottom. For research, follow the table format (Title, Takeaway, Type, Year, Citations, DOI) and sort by citations. |
There was a problem hiding this comment.
To ensure consistency with the instructions in CLAUDE.md (which specifies "sort by citations descending"), consider explicitly stating "descending" here as well. This prevents contributors from sorting in ascending order.
| 3. **Keep the format.** Match the structure of the file you're editing. For regulatory and market files, keep the "last updated" date at the bottom. For research, follow the table format (Title, Takeaway, Type, Year, Citations, DOI) and sort by citations. | |
| 3. **Keep the format.** Match the structure of the file you're editing. For regulatory and market files, keep the "last updated" date at the bottom. For research, follow the table format (Title, Takeaway, Type, Year, Citations, DOI) and sort by citations descending. |
|
|
||
| **Use blind references and sharper hiring questions.** Ask "On a scale of 1–10, how likely are you to rehire this person?" and "Were they in the top 1% or top 10% of your reports?" Founders overrate their own gut, and roughly half of senior hires are gone within 18 months. Shrink the interview panel (HubSpot went from 8 to 4) and pick spiky candidates with real strengths over safe 3-out-of-4 generalists. A mis-hire on a small team is slow and expensive to correct. — Brian Halligan, *Lenny's Podcast* | ||
|
|
||
| **Agree on what the goal word means before anyone builds.** Forsgren's most common failure — 80% of teams — is starting work before defining the goal. "Improve developer experience" splinters into culture versus tooling versus friction, and teams spend months building the wrong thing; the fix costs one week of writing it down. When a district says it wants to "improve engagement" or "use AI," pin the exact definition in writing before scoping the pilot, or you'll ship something they didn't mean and lose the renewal. — Nicole Forsgren, *Lenny's Podcast* |
There was a problem hiding this comment.
The phrase "goal word" in the header **Agree on what the goal word means before anyone builds.** seems to be a typo or awkward phrasing. It would be clearer and more professional to use "goal" or "goal's definition" instead.
| **Agree on what the goal word means before anyone builds.** Forsgren's most common failure — 80% of teams — is starting work before defining the goal. "Improve developer experience" splinters into culture versus tooling versus friction, and teams spend months building the wrong thing; the fix costs one week of writing it down. When a district says it wants to "improve engagement" or "use AI," pin the exact definition in writing before scoping the pilot, or you'll ship something they didn't mean and lose the renewal. — Nicole Forsgren, *Lenny's Podcast* | |
| **Agree on what the goal means before anyone builds.** Forsgren's most common failure — 80% of teams — is starting work before defining the goal. "Improve developer experience" splinters into culture versus tooling versus friction, and teams spend months building the wrong thing; the fix costs one week of writing it down. When a district says it wants to "improve engagement" or "use AI," pin the exact definition in writing before scoping the pilot, or you'll ship something they didn't mean and lose the renewal. — Nicole Forsgren, *Lenny's Podcast* |
|
|
||
| **Count your "barrels," not your headcount.** Barrels are the rare people who own an initiative end-to-end and get it over the hill; everyone else is ammunition that amplifies a barrel. PayPal had only 12–17 barrels among 254 people. Adding ammunition behind the same barrels adds coordination tax without throughput, so staff against your number of true owners rather than raising and over-hiring against a thin pilot or sales motion. — Keith Rabois, *Lenny's Podcast* | ||
|
|
||
| **AI adoption needs top-down budget plus a bottom-up tiger team.** Pure mandates fail; what works pairs exec buy-in with a small team of your most excited people who find real workflows, run hackathons, and evangelize — often technical-adjacent non-engineers like the ops lead who's an Excel wizard, not coders. This is exactly how your district and university customers will or won't succeed, so build a customer-side champion team into your implementation plan rather than relying on an administrator's mandate teachers quietly ignore. — Sherwin Wu, *Lenny's Podcast* |
There was a problem hiding this comment.
A couple of minor grammatical improvements can enhance the readability of this lesson:
- Change "technical-adjacent" to "technically-adjacent" (adverb modifying adjective).
- Add "that" before "teachers quietly ignore" to make the relative clause clearer.
| **AI adoption needs top-down budget plus a bottom-up tiger team.** Pure mandates fail; what works pairs exec buy-in with a small team of your most excited people who find real workflows, run hackathons, and evangelize — often technical-adjacent non-engineers like the ops lead who's an Excel wizard, not coders. This is exactly how your district and university customers will or won't succeed, so build a customer-side champion team into your implementation plan rather than relying on an administrator's mandate teachers quietly ignore. — Sherwin Wu, *Lenny's Podcast* | |
| **AI adoption needs top-down budget plus a bottom-up tiger team.** Pure mandates fail; what works pairs exec buy-in with a small team of your most excited people who find real workflows, run hackathons, and evangelize — often technically-adjacent non-engineers like the ops lead who's an Excel wizard, not coders. This is exactly how your district and university customers will or won't succeed, so build a customer-side champion team into your implementation plan rather than relying on an administrator's mandate that teachers quietly ignore. — Sherwin Wu, *Lenny's Podcast* |
|
|
||
| **Raising the price often raises signups in B2B.** Price signals quality and selects which buyers even consider you. One founder selling to enterprise and government at $300/year saw zero change after switching to $300/month — 12x — proof he was nowhere near the ceiling. A cheap per-teacher price screens you out of district procurement, where buyers equate low price with weak security, support, and governance. — Jason Cohen, *Lenny's Podcast* | ||
|
|
||
| **Reposition from savings to growth and charge far more.** Pitched as "cut your AdWords cost in half," a tool is worth ~$5K/month because the buyer keeps most of the savings; pitched as "double your leads at the same ROI," the identical tool is worth ~$40K, because growth budget dwarfs cost-cutting budget. Sell "raise completion, retention, or enrollment," not "save teachers time" — outcomes leadership is measured on carry an order-of-magnitude bigger budget than efficiency claims. — Jason Cohen, *Lenny's Podcast* |
There was a problem hiding this comment.
The clause "outcomes leadership is measured on carry..." is slightly difficult to parse. Adding the relative pronoun "that" (i.e., "outcomes that leadership is measured on") would significantly improve readability.
| **Reposition from savings to growth and charge far more.** Pitched as "cut your AdWords cost in half," a tool is worth ~$5K/month because the buyer keeps most of the savings; pitched as "double your leads at the same ROI," the identical tool is worth ~$40K, because growth budget dwarfs cost-cutting budget. Sell "raise completion, retention, or enrollment," not "save teachers time" — outcomes leadership is measured on carry an order-of-magnitude bigger budget than efficiency claims. — Jason Cohen, *Lenny's Podcast* | |
| **Reposition from savings to growth and charge far more.** Pitched as "cut your AdWords cost in half," a tool is worth ~$5K/month because the buyer keeps most of the savings; pitched as "double your leads at the same ROI," the identical tool is worth ~$40K, because growth budget dwarfs cost-cutting budget. Sell "raise completion, retention, or enrollment," not "save teachers time" — outcomes that leadership is measured on carry an order-of-magnitude bigger budget than efficiency claims. — Jason Cohen, *Lenny's Podcast* |
|
|
||
| **Keep the friction that helps a user see the product is for them.** Don't reflexively strip onboarding steps. Anthropic, MasterClass, Mercury, and Calm all keep deliberately long onboarding quizzes, because asking who the user is lets you route them to the right feature and personalize later. Cut friction that adds nothing; keep friction that earns a better first experience, and reuse the profile data for re-engagement. Activation in edtech hinges on a teacher or learner seeing "this is for my subject, grade, or role." — Amol Avasare, *Lenny's Podcast* | ||
|
|
||
| **Protect your notification and email channels as a hard constraint.** Groupon escalated to five emails a day, briefly won on metrics, then permanently lost the channel, because opted-out users never come back. Duolingo let the team optimize timing, copy, and images freely but required CEO approval to raise frequency. Edtech leans on reminder emails and push to drive student and teacher engagement, so over-testing frequency can burn the exact channel that keeps a district's usage numbers up at renewal time. — Jorge Mazal, [*Lenny's Newsletter*](https://www.lennysnewsletter.com/p/how-duolingo-reignited-user-growth) |
There was a problem hiding this comment.
Using "push notifications" instead of just "push" is more precise and professional in this context.
| **Protect your notification and email channels as a hard constraint.** Groupon escalated to five emails a day, briefly won on metrics, then permanently lost the channel, because opted-out users never come back. Duolingo let the team optimize timing, copy, and images freely but required CEO approval to raise frequency. Edtech leans on reminder emails and push to drive student and teacher engagement, so over-testing frequency can burn the exact channel that keeps a district's usage numbers up at renewal time. — Jorge Mazal, [*Lenny's Newsletter*](https://www.lennysnewsletter.com/p/how-duolingo-reignited-user-growth) | |
| **Protect your notification and email channels as a hard constraint.** Groupon escalated to five emails a day, briefly won on metrics, then permanently lost the channel, because opted-out users never come back. Duolingo let the team optimize timing, copy, and images freely but required CEO approval to raise frequency. Edtech leans on reminder emails and push notifications to drive student and teacher engagement, so over-testing frequency can burn the exact channel that keeps a district's usage numbers up at renewal time. — Jorge Mazal, [*Lenny's Newsletter*](https://www.lennysnewsletter.com/p/how-duolingo-reignited-user-growth) |
|
|
||
| **Make the first call a working session that leaves an asset.** Stripe's first sales call was a whiteboard of the customer's own payments architecture, so the buyer walked away with a diagram they'd never drawn — helped, not quizzed. Run your first school meeting as a co-mapping of their student journey, data flow, or current tool stack, so they leave with something useful even if they don't buy. That's the trust that wins a slow committee sale. — Jeanne DeWitt Grosser, *Lenny's Podcast* | ||
|
|
||
| **Ask spicier discovery questions.** "What's top of mind?" has gone generic and yields rehearsed answers. Fain asks emotion-surfacing questions instead: "What's the most urgent thing you're scared of messing up?" and "What is your board pushing you on?" With a district or L&D buyer, probe what their board, state accountability metrics, or superintendent are pressuring them on this year, then tie your tool to relieving that specific fear. — Jessica Fain, *Lenny's Podcast* |
There was a problem hiding this comment.
Subject-verb agreement issue: "their board, state accountability metrics, or superintendent are pressuring them on...". Since the subjects are joined by "or", the verb should agree with the nearest subject ("superintendent", which is singular, requiring "is").
To keep the plural verb "are" and make the sentence flow naturally, you can reorder the list so that the plural subject ("state accountability metrics") is closest to the verb.
| **Ask spicier discovery questions.** "What's top of mind?" has gone generic and yields rehearsed answers. Fain asks emotion-surfacing questions instead: "What's the most urgent thing you're scared of messing up?" and "What is your board pushing you on?" With a district or L&D buyer, probe what their board, state accountability metrics, or superintendent are pressuring them on this year, then tie your tool to relieving that specific fear. — Jessica Fain, *Lenny's Podcast* | |
| **Ask spicier discovery questions.** "What's top of mind?" has gone generic and yields rehearsed answers. Fain asks emotion-surfacing questions instead: "What's the most urgent thing you're scared of messing up?" and "What is your board pushing you on?" With a district or L&D buyer, probe what their board, superintendent, or state accountability metrics are pressuring them on this year, then tie your tool to relieving that specific fear. — Jessica Fain, *Lenny's Podcast* |
- CONTRIBUTING: "sort by citations" -> "sort by citations descending" (matches CLAUDE.md) - operator-lessons: "goal word" -> "goal"; add missing "that" in two clauses; "push" -> "push notifications"; fix subject-verb order in the discovery-questions lesson Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Restores the v1.4.0 count-agnostic documentation policy: README, ARCHITECTURE,
CLAUDE.md, and the social-preview card no longer hardcode "376 papers" / "71
lessons" / "15 jobs". Generic phrasing ("hundreds of", "dozens of") instead;
the precise, current counts live only in data/research/README.md, which updates
as the corpus grows. The social card now shows the knowledge pillars, not numbers.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request pivots the EdTech Founder Stack from an interactive set of AI skills to a curated, AI-friendly markdown knowledge base. It removes the skills/ directory and the setup script, introduces a new data/operator-lessons.md file with distilled field lessons, and updates all documentation and metadata files to reflect this change. The review feedback correctly points out that the new, longer stat labels in docs/social-preview.html might wrap awkwardly or overflow the fixed-width card layout, and suggests more concise alternatives.
| <div class="stat-number">Research</div> | ||
| <div class="stat-label">Peer-reviewed learning science</div> | ||
| </div> | ||
| <div class="stat"> | ||
| <div class="stat-number">376</div> | ||
| <div class="stat-label">Research Papers</div> | ||
| <div class="stat-number">Playbooks</div> | ||
| <div class="stat-label">Operator lessons from the field</div> | ||
| </div> | ||
| <div class="stat"> | ||
| <div class="stat-number">15</div> | ||
| <div class="stat-label">Validated Jobs</div> | ||
| <div class="stat-number">Frameworks</div> | ||
| <div class="stat-label">Higher-ed jobs & AI-native</div> | ||
| </div> |
There was a problem hiding this comment.
The new stat labels are significantly longer than the original ones (e.g., Peer-reviewed learning science is 30 characters vs. AI Skills which was 9 characters). Since .stat-label has text-transform: uppercase; letter-spacing: 1px; applied in CSS, these long labels will expand horizontally and are highly likely to wrap awkwardly or overflow the fixed 1280px width of the social preview card.
Consider using more concise labels to ensure they fit cleanly within the layout.
| <div class="stat-number">Research</div> | |
| <div class="stat-label">Peer-reviewed learning science</div> | |
| </div> | |
| <div class="stat"> | |
| <div class="stat-number">376</div> | |
| <div class="stat-label">Research Papers</div> | |
| <div class="stat-number">Playbooks</div> | |
| <div class="stat-label">Operator lessons from the field</div> | |
| </div> | |
| <div class="stat"> | |
| <div class="stat-number">15</div> | |
| <div class="stat-label">Validated Jobs</div> | |
| <div class="stat-number">Frameworks</div> | |
| <div class="stat-label">Higher-ed jobs & AI-native</div> | |
| </div> | |
| <div class="stat-number">Research</div> | |
| <div class="stat-label">Learning Science</div> | |
| </div> | |
| <div class="stat"> | |
| <div class="stat-number">Playbooks</div> | |
| <div class="stat-label">Field Lessons</div> | |
| </div> | |
| <div class="stat"> | |
| <div class="stat-number">Frameworks</div> | |
| <div class="stat-label">Jobs & AI-Native</div> | |
| </div> |
The count-agnostic labels were long enough to wrap/overflow the fixed 1280px card (uppercase + letter-spacing). Tightened to Learning Science / Field Lessons / Jobs & AI-Native. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request pivots the EdTech Founder Stack from an interactive set of CLI skills to a curated, AI-friendly markdown knowledge base. It removes the skills directory and the setup script, adds a comprehensive collection of operator lessons in data/operator-lessons.md, and updates all project documentation to reflect this new architecture. The review feedback focuses on minor grammatical, punctuation, and formatting improvements within the newly added operator lessons file.
|
|
||
| **A dead reply rate is a premise problem, not a tooling problem.** Two companies ran identical outreach and got 2% versus 12% interest; the only difference was the sharpness of the founder's insight into the problem. If cold outreach to principals or L&D leads flops, suspect your premise before blaming the email tool. "Better than your current LMS" gets ignored; a felt pain like a compliance deadline or teacher time gets replies. — Jen Abel, *Lenny's Podcast* | ||
|
|
||
| **Chase the buyer who pulls it out of your hands.** Hold a strong thesis about how the world is changing, but stay loose on the exact wedge and go after the customer who is shockingly easy to sell. If selling the next customer is a grind, the business won't scale — so trust the eager buyer over the district you spend nine months persuading that your pedagogically "correct" product is right. — Brendan Foody, *Lenny's Podcast* |
There was a problem hiding this comment.
The phrase the customer who is shockingly easy to sell should be updated to the customer who is shockingly easy to sell to for grammatical precision, as the customer is the recipient of the sale rather than the product being sold.
| **Chase the buyer who pulls it out of your hands.** Hold a strong thesis about how the world is changing, but stay loose on the exact wedge and go after the customer who is shockingly easy to sell. If selling the next customer is a grind, the business won't scale — so trust the eager buyer over the district you spend nine months persuading that your pedagogically "correct" product is right. — Brendan Foody, *Lenny's Podcast* | |
| **Chase the buyer who pulls it out of your hands.** Hold a strong thesis about how the world is changing, but stay loose on the exact wedge and go after the customer who is shockingly easy to sell to. If selling the next customer is a grind, the business won't scale — so trust the eager buyer over the district you spend nine months persuading that your pedagogically "correct" product is right. — Brendan Foody, *Lenny's Podcast* |
|
|
||
| **Own the data or be the model — the middle gets crushed.** Point solutions lack the first-party data to do useful joins, and integrating via flat-file feeds is drinking through a straw. Defensible AI products either own the data ("the mine") or provide the model ("the shovels"); renting both crushes your unit economics. A single-feature tool pulling SIS/LMS/assessment data through brittle integrations won't be allowed good economics by the platforms it depends on — owning a data wedge and accumulating each teacher's students, materials, and corrections is the moat. — Matt MacInnis, *Lenny's Podcast* | ||
|
|
||
| **Distrust one-click agents and multi-agent "swarms."** Replacing a real workflow takes roughly four to six months even with good data, because enterprise data is messy — duplicate functions, broken taxonomies, undocumented rules. Current models can't reliably run a utopia of peer agents that self-coordinate; what works is one supervisor agent (or a human) orchestrating sub-agents. School SIS, LMS, and roster data are exactly this messy, so budget months not weeks — and don't architect a tutoring product as a swarm of specialists talking to each other when a student or parent is on the other end. — Aishwarya Naresh Reganti & Kiriti Badam, *Lenny's Podcast* |
There was a problem hiding this comment.
Adding a comma in budget months not weeks to make it budget months, not weeks would improve readability.
| **Distrust one-click agents and multi-agent "swarms."** Replacing a real workflow takes roughly four to six months even with good data, because enterprise data is messy — duplicate functions, broken taxonomies, undocumented rules. Current models can't reliably run a utopia of peer agents that self-coordinate; what works is one supervisor agent (or a human) orchestrating sub-agents. School SIS, LMS, and roster data are exactly this messy, so budget months not weeks — and don't architect a tutoring product as a swarm of specialists talking to each other when a student or parent is on the other end. — Aishwarya Naresh Reganti & Kiriti Badam, *Lenny's Podcast* | |
| **Distrust one-click agents and multi-agent "swarms."** Replacing a real workflow takes roughly four to six months even with good data, because enterprise data is messy — duplicate functions, broken taxonomies, undocumented rules. Current models can't reliably run a utopia of peer agents that self-coordinate; what works is one supervisor agent (or a human) orchestrating sub-agents. School SIS, LMS, and roster data are exactly this messy, so budget months, not weeks — and don't architect a tutoring product as a swarm of specialists talking to each other when a student or parent is on the other end. — Aishwarya Naresh Reganti & Kiriti Badam, *Lenny's Podcast* |
|
|
||
| **Count your "barrels," not your headcount.** Barrels are the rare people who own an initiative end-to-end and get it over the hill; everyone else is ammunition that amplifies a barrel. PayPal had only 12–17 barrels among 254 people. Adding ammunition behind the same barrels adds coordination tax without throughput, so staff against your number of true owners rather than raising and over-hiring against a thin pilot or sales motion. — Keith Rabois, *Lenny's Podcast* | ||
|
|
||
| **AI adoption needs top-down budget plus a bottom-up tiger team.** Pure mandates fail; what works pairs exec buy-in with a small team of your most excited people who find real workflows, run hackathons, and evangelize — often technical-adjacent non-engineers like the ops lead who's an Excel wizard, not coders. This is exactly how your district and university customers will or won't succeed, so build a customer-side champion team into your implementation plan rather than relying on an administrator's mandate that teachers quietly ignore. — Sherwin Wu, *Lenny's Podcast* |
There was a problem hiding this comment.
The term technical-adjacent should be updated to technically adjacent (using the adverb form) for better grammatical flow.
| **AI adoption needs top-down budget plus a bottom-up tiger team.** Pure mandates fail; what works pairs exec buy-in with a small team of your most excited people who find real workflows, run hackathons, and evangelize — often technical-adjacent non-engineers like the ops lead who's an Excel wizard, not coders. This is exactly how your district and university customers will or won't succeed, so build a customer-side champion team into your implementation plan rather than relying on an administrator's mandate that teachers quietly ignore. — Sherwin Wu, *Lenny's Podcast* | |
| **AI adoption needs top-down budget plus a bottom-up tiger team.** Pure mandates fail; what works pairs exec buy-in with a small team of your most excited people who find real workflows, run hackathons, and evangelize — often technically adjacent non-engineers like the ops lead who's an Excel wizard, not coders. This is exactly how your district and university customers will or won't succeed, so build a customer-side champion team into your implementation plan rather than relying on an administrator's mandate that teachers quietly ignore. — Sherwin Wu, *Lenny's Podcast* |
|
|
||
| --- | ||
|
|
||
| *Distilled from the free public starter dataset of Lenny's Podcast and Lenny's Newsletter (github.com/LennysNewsletter/lennys-newsletterpodcastdata), used under its personal/non-commercial terms. Takeaways are paraphrased, not quoted. Last updated 2026-05-29.* |
There was a problem hiding this comment.
The GitHub repository reference github.com/LennysNewsletter/lennys-newsletterpodcastdata is currently plain text. It would be better to format it as a clickable markdown link to match the other links in the document.
| *Distilled from the free public starter dataset of Lenny's Podcast and Lenny's Newsletter (github.com/LennysNewsletter/lennys-newsletterpodcastdata), used under its personal/non-commercial terms. Takeaways are paraphrased, not quoted. Last updated 2026-05-29.* | |
| *Distilled from the free public starter dataset of Lenny's Podcast and Lenny's Newsletter ([github.com/LennysNewsletter/lennys-newsletterpodcastdata](https://github.com/LennysNewsletter/lennys-newsletterpodcastdata)), used under its personal/non-commercial terms. Takeaways are paraphrased, not quoted. Last updated 2026-05-29.* |
- "easy to sell" -> "easy to sell to" - "budget months not weeks" -> "budget months, not weeks" - "technical-adjacent" -> "technically adjacent" - footer: make the source repo URL a clickable markdown link Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
1 participant
Summary
Repositions EdTech Founder Stack from an interactive-skills toolkit into a curated, AI-friendly markdown knowledge base for edtech founders. The knowledge in
data/is now the product: point Claude Code, Cursor, or ChatGPT at it, or read it directly.What's new
data/operator-lessons.md— 71 field lessons across validation, product, building with AI, pilots, go-to-market, sales, pricing, growth & retention, fundraising, and team. Distilled and attributed from the free public archive of Lenny's Podcast and Lenny's Newsletter, mapped to selling into schools, universities, and L&D. Paraphrased (not quoted) under the source's personal/non-commercial terms, with verified episode links where available.What changed
setupscript are gone. PR/issue templates, CODEOWNERS, and the social-preview card updated to match.The four knowledge layers
data/research/)data/)data/operator-lessons.mdHeads up: CI will be red
.github/workflows/validate.ymlstill checks for the now-removedskills/. Updating a workflow file needs theghtoken'sworkflowscope, which isn't granted yet, so that one file is intentionally left unchanged in this branch. The failing check is cosmetic; the CI cleanup lands the moment the scope is added (gh auth refresh -h github.com -s workflow).🤖 Generated with Claude Code