How QWF Learns

How QWF Learns

A living reference of the sources, methods, and disciplines Quietly Working Foundation uses to stay ahead… built for owner-operators and Executive Directors who don’t have a research team.

This document updates constantly. Last substantial revision: 2026-04-21. If something here is dead or drifting, tell us and we’ll prune it.


Why This Exists

Someone asked a great question on the TechSoup community forum this week. Besides TechSoup webinars… what YouTube channels, newsletters, blogs, and podcasts do you actually use to learn about emerging tech?

Simple question. Answer that deserves more than a forum reply.

So this is the real answer. The full stack. The sources we trust, the method we use to find them, the discipline we use to prune them, and the principle we use when we’re about to get sucked into the next shiny distraction.

This is written for one specific person… the one in daily operations. The Executive Director who’s also the IT person who’s also the grant writer. The small business owner still fielding customer emails at 9pm. The nonprofit founder wearing four hats on a Tuesday.

You are the learning function in your org. Nobody’s handing you a curated research digest. Here’s how we do it at QWF, and here’s every source we’d recommend if you asked.

If you haven’t seen it yet, this article lives alongside a few others that make more sense together than apart… our QWU Values (why we publish any of this), How to Give Your AI Agent Superpowers (the architecture that makes the learning compound), Nonprofit Tech Access Guide (how nonprofits actually get software discounts worth learning about), and The Tool Shed (what we actually run). Read them in any order. They all feed each other.


The Method First… Then The Sources

The list of sources is long. You’ll see it. But the list isn’t the answer.

The answer is the method you use to keep a list like this from drowning you.

Three truths we learned the hard way:

1. Subtract more than you add.

In 2026, restraint is the status signal. Subscribing to 40 newsletters makes you slower, not smarter. Every quarter, put a calendar event titled “Subscription Audit” and ask each source: has this actually changed a decision I made in the last 90 days? If no… cut it. New subscription only allowed if an old one retires. One in, one out.

Professional content curators find they can cut 30-40% of sources without losing any useful signal. Bet you can too.

2. Follow builders, not commentators.

The highest-signal creators are practitioners who write as a byproduct of the work. They post less. They miss news cycles. Their takes age well because they have to.

Andrej Karpathy ships nanoGPT on a weekend and teaches you more about LLMs than a year of AI newsletters. Patrick McKenzie runs four companies and pre-empts mainstream fintech coverage by 18 months in Bits about Money. Beth Kanter has been writing credibly about nonprofit tech since before the smartphone existed and still isn’t hyping it.

For every commentator in your feed, add two practitioners. Practitioners ship monthly instead of daily. That’s the feature, not the bug.

3. A few sources understood deeply beats many sources skimmed.

Warren Buffett reads 500 pages a week and has for 60 years. Naval Ravikant treats books like blogs… skims widely, dives deep on interesting parts, no guilt about not finishing. Morgan Housel reads 1 to 2 books a month and re-reads the ones that hit.

Pick 3-5 sources you’ll read for the next decade. Re-read them. Highlight them. Annotate them. Everything else is rotation.


The Curation Filters We Run Every Source Through

Before a source gets into our stack, it has to survive four questions.

Did they predict things that came true? If they made big calls in 2022 or 2023, did any of them land? Honest tracking… not cherry-picked hits.

Are they the builder or the talker? Do they ship the thing they describe, or do they only describe the things others ship? Both roles are valuable. But if you only consume commentators, you get further from reality with every newsletter.

Is it first-hand or secondary? A Substack summarizing a paper you could’ve read yourself is worse than the paper. A thread summarizing a Substack post is worse than the Substack post. Cut middle layers ruthlessly.

Does it respect your time? Consistency over cadence. Fewer posts with higher hit-rate beats daily volume every time. If the author writes weekly and it’s always a slog to get through, unsubscribe without guilt.


How QWF Actually Finds New Signals

The discovery process matters as much as the subscriptions.

The research YouTube account. We keep a Google account used only for professional discovery. Never watch entertainment on it. The algorithm becomes a curation tool instead of a distraction loop.

The comment section scan. When a creator we trust posts, we scan the comments for other creators they engage with. Builders tag each other. Fans make surprisingly good recommendations.

Newsletter network effects. Substack’s recommendation engine genuinely works at this point. One good newsletter leads to three adjacent good ones. Start with one you love and mine the “recommended” list.

Community mining. Private communities are where real recommendations live. Not the $5K-a-year ones necessarily… sometimes a free Discord with 80 working founders beats a flashy paid network of 80,000 aspirants. The bar for inclusion is the product.

Reverse chronological check-ins. Every 90 days we check what we were reading 6 months ago. Sources that still hit… keep. Sources we forgot we subscribed to… cut. No sentiment.


The Stack (Organized by What It’s For)

Everything below was verified as actively publishing within the last 60 days as of April 2026. Filtered for signal, practitioner trust, and accessibility. Paid sources marked clearly… most of this is free.

For Nonprofit IT Specifically

This is the answer to Chris’s actual question. If your job title includes “IT” or “Operations” at a nonprofit, start here.

SourceFormatWhy It Earns Its Slot
NTEN ConnectMonthly newsletter + annual conferenceThe de facto professional association for nonprofit technologists. Vendor-agnostic.
Community IT Innovators PodcastWeekly podcast + monthly free webinarsHands-down the most IT-practical source we’ve found. Microsoft 365, cybersecurity, cloud, AI policy… all contextualized for small nonprofits.
Nonprofit Tech for GoodWeekly newsletter + affordable certificate programsHeather Mansfield’s 15+ year track record. Digital fundraising and social media done right.
Beth Kanter’s BlogBlog + “All Things AI” newsletterThe sector’s most credible voice on generative AI adoption without hype. Co-authored The Smart Nonprofit with Allison Fine.
The Chronicle of PhilanthropyDaily + print~350,000 nonprofit pros subscribe. Call it the Wall Street Journal of the nonprofit world.
Candid InsightsBlog + newsCandid runs the 990 data infrastructure the sector runs on.
Tech ImpactBlog + workforce programsRuns actual IT workforce training. Pragmatic, operator perspective.

For Emerging Tech and AI (Practitioner-Grade)

These are the sources that help you understand what’s happening and what it means for your operation. Not the hype layer… the signal layer.

Newsletters worth the inbox space:

SourceCostWhy
One Useful Thing by Ethan MollickFreeSingle best source for operationalizing AI this month. Every essay includes experiments you can run. Wharton professor. No hype.
Stratechery by Ben ThompsonFree weekly + $150/yr dailyThe most-cited tech analyst among operators. 12+ years of consistency. Strategic framing beats news.
Exponential View by Azeem AzharFree + paidThe 3-5 year horizon. Cited by HBR and the FT.
Import AI by Jack ClarkFreeWeekly. Co-founder of Anthropic writing at the frontier but still explaining for outsiders.
The Pragmatic Engineer by Gergely Orosz~$15/mo500K+ readers. The “what’s really happening inside tech companies” view even non-engineers benefit from.
The Batch by Andrew NgFreeWeekly ML research and industry summary. Andrew Ng’s credibility is unmatched in ML education.
Latent Space by swyxFreeWritten for AI engineers by AI engineers. Highest signal for technical practitioners.
The Rundown AIFree2M+ subscribers. Daily 5-minute breadth. Pair with One Useful Thing for depth.

YouTube channels worth subscribing to:

ChannelWhat You Get
Andrej KarpathyThe gold standard. “Zero to Hero” series teaches neural networks from scratch. Former Tesla AI director, OpenAI founding member.
AI ExplainedWeekly research paper breakdowns. Calm, measured, researcher-aware.
Two Minute Papers5-10 minute summaries of AI research. Dr. Károly Zsolnai-Fehér at TU Wien.
Matt WolfeAI tool reviews and weekly news. Entrepreneur lens… focuses on what you can use today.
Wes RothDaily AI news and agent developments. Catches stories before mainstream press.
Nate B. JonesStrategic AI implications for product and operations leaders. 20-year product veteran.
Cleo AbramOptimistic tech explainers. Rigorous sourcing, no doomscroll.

Podcasts worth the commute:

PodcastWhy
Dwarkesh PodcastMost pointed AI interviewer working. Extracts specifics where others get vague answers.
Hard Fork (NYT)Kevin Roose and Casey Newton. Skeptical but not cynical. Accessible for non-technical EDs.
Last Week in AIResearcher hosts. Structural weekly coverage across research, industry, policy, ethics.
No PriorsSarah Guo and Elad Gil interviewing AI founders, often pre-mainstream.
Latent SpaceBest source on open-source AI specifically.

For Operators and Founders in Daily Ops

These are the sources that teach you how to run the thing once you’ve figured out what to build.

SourceFormatWhy
Lenny’s Newsletter + PodcastFree + paidFormer Airbnb PM interviewing operators. Unusually candid for the genre.
Every.toDaily bundle + podcastDan Shipper’s Chain of Thought column is the flagship. Operator-authentic… the team ships AI products and writes about shipping them.
Not Boring by Packy McCormickFree weeklyLong-form strategy essays. Former lawyer, so explanations are clear. 250K+ subscribers.
First Round ReviewFreeTactical operator playbooks on hiring, comp, management. Drawn from portfolio companies. No sponsored content.
My First MillionPodcastSam Parr (built and sold The Hustle) and Shaan Puri. Trend-spotting and marketing tactics.
AcquiredPodcast3-4 hour deep dives on single companies. Best strategy history you can get for free.

For System Design and Engineering Context

Chris mentioned ByteByteGo… here’s the wider set. Even if you’re not shipping code, knowing how modern systems work makes you a better buyer.

SourceWhy
ByteByteGoSystem design made visual. Alex Xu is exceptional at explaining scale.
Netflix TechBlogData pipelines, streaming infrastructure.
Stripe EngineeringPayment systems, API design. Industry gold standard.
Cloudflare BlogNetworking, edge compute, security. Unusually readable.
GitHub EngineeringGit internals, Copilot infrastructure.
TLDR NewsletterDaily 5-minute tech news digest. 750K+ readers.

For Creative and Content Production

QWF has a creative arm (Missing Pixel, QQT, the Transparency Project). If you make content in 2026, these are the sources we watch.

SourceWhy
Corridor CrewWorking VFX artists showing actual tools and actual mistakes. AI film experiments that ship.
Curious RefugeRuns the largest AI-film community. Hosts festivals Hollywood now attends (WAiFF Cannes, AI on the Lot).
The Futur by Chris DoDesign business, brand strategy, client management. 25+ years in design leadership.
MKBHD + WaveformTurned down 7-figure crypto sponsorships to protect editorial independence. Tells you what you need to know about the tech review space.
Fireship100-second programming explainers. Great “keep current” diet for small teams.

For Security and Privacy

Increasingly your job whether you want it or not. Breach reporting, practical guidance.

SourceWhy
Krebs on SecurityOriginal investigative cybercrime reporter. Often breaks stories 72 hours before mainstream.
Schneier on Security30+ years of systems-thinking security writing. Author of Applied Cryptography.
Risky BusinessRequired listening for working security pros. Skeptical, informed, funny.
SANS NewsBitesBi-weekly curated security news with expert commentary. Free. Great for small-org CISOs.
Smashing Security by Graham CluleyAccessible security news great for non-technical staff training.

For Leadership and Meaning

Selectively curated. This is TIG’s chaplain-and-ED shelf. No padding.

SourceWhy
A Bit of Optimism by Simon SinekLeadership, meaning, purpose. Spiritually open without being religious.
The Carey Nieuwhof Leadership PodcastChurch + business leadership. Interviews span Seth Godin, Adam Grant, Patrick Lencioni. Rare source that takes both faith and organizational rigor seriously.
PraxisRedemptive entrepreneurship. Building ventures from a faith perspective.
Farnam Street by Shane ParrishMental models, timeless thinking. The Knowledge Project podcast is a goldmine.

The Tools That Hold It All Together

Subscribing to sources is easy. Turning that flow into something useful is the real work.

Before I list tools… an honest warning. What follows is a three-level ladder, not a single stack. Almost every guide you read online shows you Level 1 and implies that’s all there is. Then people try it, hit hallucinations on their tenth query, and conclude AI isn’t reliable enough for their org. The AI isn’t broken. The architecture is.

QWF runs on all three levels. Most readers should start at Level 1 and stay there until it stops working. When it stops working, come back and build Level 2.

Level 1 … The Starter Kit (Where 90% Of You Should Begin)

This is genuinely enough for most solo operators and small teams for a long time.

Readwise Reader ($9.99/mo) pulls articles, PDFs, newsletters, tweets, and YouTube transcripts into one queue. Highlights sync to Obsidian automatically. Email becomes a task inbox again. Reading happens somewhere designed for it.

Obsidian (free) is where the highlights live forever. Markdown files, local-first, no cloud lock-in. When the AI hype cycle ends and tools we love today get acquired and killed, our notes will still be here.

Feedly or Inoreader (free to $6/mo) for RSS triage across many sources at speed. RSS adoption climbed 34% year-over-year in 2026. It isn’t dead… it just waited out the algorithmic-feed era. Come home.

NotebookLM (free) for topical research sessions. Dump 10 PDFs and a few transcripts, ask questions, get answers grounded only in your sources. Different beast from general ChatGPT… it won’t hallucinate outside what you fed it.

Claude Projects ($20/mo) for reasoning-heavy work. Pin your source documents, write a system prompt that encodes your priorities, query repeatedly. The second and tenth queries are where the value lives. Never paste a 90-minute transcript asking it to “just summarize.” Build a proper project and it’ll be a genuine research partner.

The Level 1 pattern: Readwise ingests. Obsidian stores forever. NotebookLM and Claude handle ephemeral reasoning. Each tool does one job.

This works beautifully for a team of one to three. It will start to strain around ten active projects and hundreds of sources. That’s when you need Level 2.

Level 2 … The Architecture Leap (When Scale Breaks the Starter Kit)

At some point the research assistant can’t hold everything. Projects contradict each other. The same question gets asked six times across six threads. Hallucinations creep in because the model has to guess what you mean by “our standard approach.”

The answer isn’t a smarter model. The answer is a three-layer architecture that separates what the AI does from what deterministic code does.

LayerWhat Lives HereWhy
Directive (what to do)SOPs in plain markdown. Every recurring process gets a directive.Humans and agents both read the same source of truth.
Orchestration (decision making)The agent itself. Reads directives, picks the right tool, asks for clarification.This is the probabilistic layer. It’s allowed to reason.
Execution (doing the work)Deterministic scripts that handle API calls, data work, file operations. Reliable, testable, fast.This layer is NOT allowed to hallucinate. It just runs.

Why it matters… if your agent is 90% accurate at any single step, chaining five steps gives you 59% success. A coin flip. You fix that not with a better model but by pushing complexity into code that always runs the same way, and keeping the agent focused on decisions.

We wrote the full guide on this separately. Read How to Give Your AI Agent Superpowers for the complete architecture, including the shell structure, the directive format, and the sidebar on why naming conventions matter more than you think.

The tooling at Level 2:

  • Claude Code ($20/mo Pro or API billing) … the CLI version of Claude that can read files, run scripts, and actually execute work on your computer. Not “Claude Projects in a terminal.” A fundamentally different beast with filesystem access and tool use.
  • n8n self-hosted (~$38/mo all-in) … your automation layer. Receives events, runs workflows, talks to everything. We documented how we self-host on Azure in Self-Hosting n8n on Azure - The Guide I Wish Existed.
  • Supabase (free to $25/mo) … your data backbone. Postgres with auth, storage, and edge functions baked in. Find them at supabase dot com.
  • Your existing Obsidian vault … now serving double duty as the human-readable surface of the system.

This is where real operators stop being “AI users” and start being AI operators.

Level 3 … The Compounding System (Where Our Actual Advantage Lives)

This is where we stop talking tools and start talking about the structured knowledge the system runs on. This level is what makes the difference between “my agent hallucinated again” and “my agent just saved me four hours.”

The Wisdom Library. 12,000+ indexed insights from every expert, podcast, paper, and video we’ve captured. Tagged by expert, topic, vertical, concern, and tool. Our agent queries this when it needs “what has the sector actually said about X?” instead of making something up. Full guide to how we built it is forthcoming… for now, the short version: every piece of content we consume gets distilled into structured rows, not just saved as highlights.

Tool Wisdom Libraries (TWLs). A dedicated directive for each tool we use heavily. 24 of them and growing. Each TWL contains the gotchas, working code examples, and “here’s what breaks and why” notes we’ve discovered in practice. When our agent needs to work with n8n or Supabase or Cloudflare Pages, it reads the TWL first. No hallucinations about API behavior because the truth is on disk.

HQ Command Center. TIG’s personal dashboard. Not a public dashboard. The single pane where decisions surface, where daily tasks live, where the Issue Tracker lives, where approval workflows route. Every other system feeds this one.

We built HQ for ourselves. And the moment we built it, we realized every working founder and Executive Director needs one. So we’re shipping a public version… Quietly Spotting. QSP gives you your own SPOT (Single Point of Truth) where every tool you already use, every data source you already pull from, and every workflow you already run feeds into one dashboard you actually control. Currently in active development and adding features weekly. If what you’re reading in this section made you lean forward, that’s your signal to go look at it.

Want to see the full picture of what QWF is building across all programs? The QWF Ecosystem widget at quietlyspotting.org is the living map. All nine apps, all the connections, no PowerPoint deck required.

The Entity Graph. 77+ expert profiles, hundreds of people and organizations in [[003 Entities]], structured the way an org chart would be. When our agent talks about “Chris Jamison,” it has a file. When it references “Ethan Mollick,” it has a file. It’s not improvising identity.

Ezer Aión. Our AI agent with name and role. Not a vendor’s chatbot. A purpose-built assistant with memory, scope, and a voice of her own. Built to run a specific subset of QWF’s work day-to-day.

The Tool Shed (public). What we use across the org, documented for our community.

This isn’t a flex. It’s the actual reason our AI stack produces real work instead of plausible-sounding fabrications. At this level, the question stops being “did the model hallucinate?” and becomes “did we miss something when we wrote the directive?”

The Honest Warning

If you try to run a real operation on Level 1 alone, you will hit hallucinations. You will conclude AI isn’t ready. You’ll tell other operators that AI isn’t ready. You’ll miss the compounding benefit because you gave up at the wrong layer.

Don’t do that.

Level 1 is a flashlight. Great tool. Built for its job. If you’re walking one person across one room, a flashlight is all you need.

Level 3 is a power grid. Different problem, different infrastructure. If you’re lighting up an entire operation, flashlights won’t scale.

Most people need the flashlight. Start there. Stay there until it actually stops working for you. Then build the ladder up.


Sources We’re Skeptical Of

Honesty is the whole point of this document. A few sources with big names that we’ve moved off or warn people about:

SourceWhy We’re Careful
a16z Podcast and FutureUseful for themes but increasingly a portfolio-marketing channel. Treat as industry PR, not journalism.
High ScalabilityHistoric archive is gold. Current cadence inconsistent. Reference only.
Most “Top AI Tools” newslettersMany are affiliate-driven. Fine as appetizer. Not main meal.
Lex Fridman (as a complete feed)Quality varies wildly by guest. Subscribe selectively, episode by episode.
LinkedIn AI influencersAlgorithmic incentive equals hype. Almost always lower signal than the same person’s Substack.
Twitter “threadbois”Threads summarizing Substack posts are never as good as the Substack posts. Skip the middle layer.

And a broader pattern worth naming… the “guru trap.” Creators who optimize for engagement over truth have tells:

  • Airtight narratives where every case study works
  • Thread-bait formatting built for screenshot virality
  • Every post funneling into a $5K cohort course
  • Reflexive contrarianism as an identity
  • No public post-mortems on their failed predictions

Before paying for any course or cohort, find a case where the creator was publicly wrong and watch how they handled it. If you can’t find one, they haven’t been building. They’ve been performing.


The Starter Stack

If you’re Chris Jamison’s profile (nonprofit IT + curious about emerging tech) and you want a minimum-viable diet, here’s the 8-item version. Three hours a week, max.

  1. NTEN Connect … community and sector
  2. Community IT Innovators Podcast … weekly IT practical
  3. One Useful Thing … practical AI
  4. Stratechery free weekly … strategic framing
  5. The Pragmatic Engineer … how tech orgs actually work
  6. Hard Fork … accessible weekly tech news
  7. Krebs on Security … security for free
  8. Beth Kanter … nonprofit AI credibility anchor

All free except Pragmatic Engineer. If you add Readwise Reader ($9.99/mo) to hold it all, your total cost is $15/month for a meaningful learning system.


The Asymmetric Bet

One more principle before we close.

The newsletters everyone reads are, by definition, priced-in knowledge. Subscribing to Nonprofit Quarterly and Stanford Social Innovation Review is table stakes. Every ED you compete with for grants reads those.

The asymmetric bet is spending $15 a month on a Substack written by someone who’s actually writing federal grants this quarter and sharing what’s working. Or a private community with 80 other EDs stress-testing AI implementations in real time.

Budget $500 to $1,000 a year for 3 to 5 paid niche sources nobody else in your sector reads.

That’s your edge.


How This Document Stays Alive

This page updates constantly. Sources get added when they earn their slot. Sources get cut when they stop earning it. Cadence is whenever… quarterly is the minimum.

If you’re reading this and a source is dead, drifting, or missing something obviously better… tell us. [email protected]. Send the link. We’ll test it and either add it or tell you why we won’t.

We share this because the question Chris asked isn’t just about newsletters. It’s about how to learn without drowning, when you’re the one who has to hold the whole operation up. We’ve been in that seat. We’re still in that seat.

Light doesn’t fight with darkness. It just shows up.

Same with information overload. You don’t beat it by arguing with the algorithm. You beat it by building a small, disciplined, trusted stack… and letting the rest pass.

We’re quietly working. Hope some of this helps.

💙


See The Whole Ecosystem

QWF builds an interconnected family of apps. Quietly Spotting is the hub. Around it orbit Quietly Writing, Quietly Quoting, Quietly Networking, Quietly Knocking, Quietly Tracking, and more. See the live ecosystem map for what’s shipped, what’s building, and how it all connects.

Everything on the transparency site is designed to interconnect. Start anywhere. Follow the links.

Acknowledgments

This list exists because people like Chris Jamison ask great questions in public. Every source on it was suggested, vetted, or proven useful by someone we trust. If your name should be on here, let us know.

Originally prompted by a question from Chris Jamison on the TechSoup community forum, April 2026.


Living document. Last updated 2026-04-21.