Each time I see one of those, followed by some "do AI" advice, I'm like Revenue per employee: OnlyFans: $37.6M PornHub: ~$8M If anything, it's not "do AI." It's "do porn." Sarcasm aside, it's comparing apples to oranges. Doesn't make sense.

"We should, in fact, subordinate development effort to the pace of experiment validation, and not vice versa." pawelbrodzinski.substack.com/p/validation...

Validation Is the New Bottleneck (Also, the Old One)

Early-stage startups have limited options for rapid validation. That makes validation timelines the bottleneck of product development.

pawelbrodzinski.substack.com

"Why don't managers use throughput/lean accounting to manage, and cost accounting to report?" -- @jrothman.com Just realized, I was doing precisely that in the late 2000s. It made intuitive sense. Even if it required translating two worlds back and forth continuously.

The latest iteration on controlling AI spend: the budgets (per team, per tool, etc.) Next up: "Why do people use *all* the budgets whether they need them or not? We should have more control over how these budgets are planned."

The useful part of user journey mapping isn't the big picture. It's finding the one part where the experience is genuinely crappy. The shitty outliers are the whole point. pawelbrodzinski.substack.com/p/user-journ...

User Journey Mapping: We Want It to Be Shit

The useful part of user journey mapping isn't the big picture. It's finding the one part where the experience is genuinely crappy.

pawelbrodzinski.substack.com

Arguably, the most socially valuable contribution of LLMs is all the pats on the back and appreciation we get in response to our prompts. So much support available for everyone. Just one prompt away. Discuss. :D

"LinkedIn was the most AI-saturated platform, where more than 40% of longform posts flagged as fully AI-generated." www.pangram.com/blog/ai-in-y...

AI Content Is Everywhere on Social Media, Especially LinkedIn | Pangram Labs

We scanned over 1 million social media posts for AI content. It turned up on every platform we checked, and 1 in 3 top LinkedIn posts flagged as AI-generated.

pangram.com

The story is: Netflix got AI to redesign the recommendation page. It increased engagement. Me: When Facebook threw away friends' content and pushed non-familiar stuff to our feeds, it did improve engagement, too. Did it make the service better, though? bit.ly/4psHAZc

GenPage: Towards End-to-End Generative Homepage Construction at Netflix

Authors: Lequn Wang, Jiangwei Pan, and Linas Baltrunas

bit.ly

I played with the statistical data about our pull requests. The gist of it? "Big just got bigger. And we get more of it, too." And that's not good news. brodzinski.com/2026/07/3x-p...

We 3.5x'd Our Pull Requests with AI: Now We Catch Fewer Bugs - Pawel Brodzinski on Leadership in Technology

AI made pull requests 3.5x bigger, but defect detection dropped. PR data from two projects shows impact of AI-generated code on quality.

brodzinski.com

When we look for a response, it’s easy to rely on an LLM, and parrot its output. That’s good enough. When we look for a thinking process, an AI response is useless. Even if it answers the question. A word-based response is (typically) general-to-specific. A thinking process is specific-to-general.

“What gets us into trouble is not what we don’t know. It’s what we know for sure that just ain’t so.” Mark Twain is a spiritual godfather of assumption mapping, my favorite product discovery technique. And yes, Twain-o-Meter is a thing. Read on. pawelbrodzinski.substack.com/p/my-favorit...

My Favorite Product Discovery Tool: Assumption Mapping

A simple technique that generates a ton of insight and can redraw the roadmap before we even start building.

pawelbrodzinski.substack.com

"As animals, human primates tend to correlate CONFIDENCE with COMPETENCE." -- @derekwwade.bsky.social

Google: "While we remain deeply committed to sustainability, reaching our climate moonshot is getting harder" Also Google: [see below]

Not a great economy alert! Wall Street Bets users think Palantir spiked today because The Onion story titled “Palantir Acquires Pentagon for $800 Billion” hit their front page. They’re guessing trading bots indexed the top post on the subreddit reddit not understanding source.

Modern startup playbook: Hypergrowth. AI. Hype. Ego. What actually improves the odds: Sustainability. Value for customers. Retention. There's hardly a better picture than the Icon (the admaker) story. pawelbrodzinski.substack.com/p/the-case-f...

The Case for Sustainability Metrics (or Don’t Be Kennan Frost)

The story of the admaker startup Icon is a warning tale for prioritizing growth over sustainability.

pawelbrodzinski.substack.com

Someone on Substack: Mildly unpopular take: half the SaaS tools I pay for, I could rebuild in a weekend with an agent and never look at a pricing tier again. Me: Question: Why have you not done that yet?

Discussion about YAGNI Me: With fewer than 5 customers, your defensive coding (addressing potentially catastrophic cases) doesn't matter. It's not code that will kill the business. Also me: That is, unless your fewer than 5 customers is a space mission. Then you definitely are gonna need it.

Consumeability vs engageability. "If you knew that an article was generated altogether by an LLM, would you write a comment? Would you engage in a discussion with a bot trying to share your experience or prove a point?" businessagilityreview.substack.com/p/engageabil...

Engageability: Why Human-Created Content Makes Sense

If quality is all that matters we shouldn't care whether a bit of content was created by human or AI. However, if we consider engaging with it, it does matter who or what is behind.

businessagilityreview.substack.com

It's really time we updated the conventional design process model to be more descriptive. We're in our third diamond era.

"Speed only helps if it is pointed in the right direction. Move fast on the wrong things and AI does not save you time. It manufactures waste at a higher rate." --Melissa Perri

When HR rebrands to People & Culture it's typically like a guinea pig. Neither Guinean nor a pig.

For decades, we learned how to have less work in progress. When it finally looked as if we internalized the idea, with AI, we're making a U-turn. It's concurrency all the way. Let's have a swarm of agents work on a thing! Gonna be good so much better!

Pay-once-use-forever used to be a smart tactic to get some early traction. These days, I consider it a smell. Too many folks with nicely looking vibe-coded stuff use that as THE business model. Get enough people to fall for that and then move on to another thing.

A neat idea: Mom-and-Pop SaaS. Except it's either misleading or an outright lie. Generating code is not remotely enough to have a SaaS-based business going on. pawelbrodzinski.substack.com/p/its-not-mo...

It’s Not Mom-and-Pop SaaS Era

The message that now anyone can build their own SaaS is misleading at best and an outright lie at worst.

pawelbrodzinski.substack.com

If and only if we, as a hiring company, care and show care, can we expect the candidates to follow suit. Otherwise, the default assumption is that it's a volume game. Something that companies that outsource their initial screening to AI completely miss.

Cold outreach: Making sure my last email didn't land in spam. Me: You've just made sure this one does.

Hiring will fall into one of two: 1. (Big one) The volume game. AI agents on both sides do the mediation. Whoever uses a better tool wins. 2. (Small one) The quality game. We design a process that requires extra effort *for both parties*. We exchange tokens of care and improve odds of good match.

AI in products is like teenage sex: everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it too. Also, I doubt we'd count teenage sex among our best intimate experiences...

How big percentage of requests to AI support agents are "I want to talk to a human" or similar?

"The Slow" don't exist. They're responding to your incentives. Slash the schedule and people hide progress and inflate estimates. The real lever is scope: how much of which goal, first? Scope is the steering wheel: newsletter.kentbeck.com/p/scope-is-t...