John M. P. Knox

@jitbit I’ll have to give it a shot again!

Just cancelled my Claude "Max" sub. Not because of the quota drama. Simply because my $20 Codex plan lasts longer than Anthropic's "Max" I spent a week throwing big refactors, new features, and weird tasks at it, including asking it to analyze Tailwind’s output.css for optimization opportunities. Final test: launching a huge marketing pipeline with multiple GSC MCP calls and @coreyhainesco's skills. It never dropped below 45% of the weekly limit. Not sure how thats possible. My guess: part of it is Claude Code system prompt being ~27k tokens, versus ~13k in Codex. P.S. Codex VS Code extension also feels more polished: nicer UI, better diffs, revert/reapply, etc.

A mental shift surfaced at my AI roundtable last month: the cost of failure with AI is almost zero. If it does something wrong, roll it back. What did you waste? Five minutes and some tokens. The old instinct — plan the prompt carefully, describe every requirement, think through the edge cases — is a holdover from a world where redoing work was expensive. Software is disposable now. One SaaS founder at the table had vibe-coded a pricing test *management system* — not a single A/B test, the whole infrastructure for running them. A couple of sentences in, result out, iterate fast. The tradeoff: speed of creation now outpaces speed of adoption. Another founder at the same table admitted to "a lot of dropped projects" — things AI spun up quickly that still need human effort to deploy and maintain. Full roundtable writeup — including AI-powered voice demos for self-serve SaaS, the "projects vertical, skills horizontal" framework, and why the founders pulling ahead are building skills that persist: movingavg.com/essays/ai-skil… #AIStrategy #FounderLife #ClaudeCode

@ray_deck 🤣

@WindAddict I want to see windaddictMCP! John as a super high end tool call.

How to put your thumb on the ai scale

Power utilities cutting off residential service seems like a potential positive feedback loop for renewable energy. Residents scramble for generation capacity, growing the renewables industry. Data centers, seeing these cheaper and better options, also build out renewables, reducing their reliance on traditional utilities. Eventually, traditional central generators become a niche business. apple.news/Asn4Ss5Y2RRCYl…

Somebody designs a manifold for a rocket and labels it "manifold." If their CAD tool quietly trained on that input, a competitor types "I need a rocket manifold" and gets a starting point that looks suspiciously familiar. That's the weirder version of shadow AI: it comes from your vendors, not your employees. JIRA added AI to Jira Cloud without asking anybody. Some CAD tools have added AI features that train on user input. The vendor pushes an update, you install it, and suddenly a feature inside an existing license is doing something with your data. As one attendee at my AI roundtable put it: "Is this actually making the product better, or is this just your way of adding AI so you're keeping up with the Joneses?" The fix is quarterly tool audits. Have your CTO inventory every tool in the organization and check whether its AI features protect your data or expose it. The catch: you can't audit what you don't know about. Full essay — including the desire-paths approach to legitimizing employee shadow usage, vendor AI quietly training on your work, and the credential firewall pattern: movingavg.com/essays/shadow-… #ShadowAI #AIGovernance #FounderLife

@ryancarson I agree, although I’m pretty sure there’s still an engineering organization in the future.

GitLab announced a layoff today. Please take this seriously. There will be many, many more. Your assignment is clear: Get skilled with agents and practice shipping to prod. It doesn't matter if you're HR, eng, infra, customer success, admin, ops, sales, whatever. As a Founder/CEO, I can tell you that I won't be hiring any employees who aren't really skilled with agents and able to ship to prod. I'm not alone in this. There is no 'engineering' org in the future.

Nostalgia con men are terrible. “Remember when you were a child and life was easy? That’s not because you were a child. It’s because Bezos had less money.” Yeah, maybe billionaires don’t pay their fair share of taxes. Maybe. But that’s not a failure of billionaires. Too much government spending is not a failure of billionaires. Bad government policy is not a failure of billionaires. These problems all start with politicians.

Truly stupid people are everywhere these days. Pretty much all Americans are vastly better off today than in 1980. It’s not even particularly close. But that doesn’t fit the narrative (“wealth is always stolen, never created”)

@jitbit There used to be incredible native apps that fit in 600kb. I bet in some cases, they ran faster than these websites too. 🤣

Claude's is 600kb. darn, the web became disgusting

Nah, we’ve been through these technology revolutions before. You’re right that demand will explode for everything downstream of AI use. We need to keep our eyes on the long term, while identifying opportunities in the short term. If you’re downstream, you better think through your pricing and business model.

AI danger is not slop, spam. bad code, fake videos etc, that can be dealt with it’s that if it makes everything 2000% more productive - you’re basically removing a dam. And removing a dam causes mare stress for the rest of the systems downstream. We’re all f*cked

A pattern with technical founders right now: they're 40 to 80 hours a week deep into AI, while their team uses it lightly or not at all. The CEO's current project is something like figuring out how to use AI for SEO. You're the CEO. Why are you doing the SEO work? You have a team of engineers. Give that problem to them — let them use AI to figure it out. Your job is strategy, go-to-market, whatever the actual constraint on the business is right now. It's not implementing SEO with AI. This is the second face of the operating-system gap: the CEO who hasn't mapped their own role, so they end up doing whatever the AI makes easy instead of whatever the business needs. AI didn't create this problem — it's a classic founder trap — but AI puts it on rocket fuel. You can suddenly produce output in any function and feel productive while the strategic work goes untouched. Full essay — including the intern test for AI permissions, why the value compounds in your context (not the model), and pushing back as the most important AI-era skill: movingavg.com/essays/map-ope… #AIStrategy #FounderLife #Leadership

A founder at this week's roundtable runs a habitual one-on-one question with his team: What could you not use AI for this week? The answers are revealing. They live in the comfortable corners of someone's week — the things they've always done by hand because they always did them by hand. A typical answer: "I had to pull out Postman and write some API requests." His follow-up: Claude could write a Python script that did 5 million API requests and tell you what it found. The exercise surfaces mental blocks rather than technical limits. And it makes the manager an AI coach, helping each team member find the AI uses inside their own workflow. Full roundtable writeup — including the fight-fire-with-fire pattern of bots reviewing bots, the IKEA effect with skills, and why token-maxing is the new lines of code: movingavg.com/essays/ai-laye… #AIStrategy #FounderLife #Leadership

@ramit Ramit, who chooses to rent. 🤣 This is almost as good a moniker as “he who must not be named.”

lol

CEOs are considering building their own CRMs because the commercially available systems don’t have adequate APIs to support the #AI agents they’re building. The current hype might frame this as “vibe coding,” but it’s not. This is engineering. Salesforce or Hubspot might have product-market fit for a 100% human sales team, but a cybernetic sales team needs a fast API with access to every feature. Agents aren’t using a GUI, and they’re not limited to touching 200 prospects a day. How many other businesses will be disrupted because their customers now expect 10X the capacity and 100% API coverage?

I think we need more of these...

At this week's AI roundtable, two founders independently described the same pattern: they've built AI tools to review their AI's output. One runs a code-review bot named Auto — a Git user, a required reviewer on every pull request, running on a 15-minute cron. By the time a human looks at a PR, Auto has already been through it. Either the developer fixed the flagged issues or modified the reviewer skill itself with a justification. Code quality has gotten a lot better. Another team built a Ruby gem called Wall-E — named for the Pixar garbage collector. It runs a weekly audit on the codebase, pairs deterministic linters with an LLM triage step, and opens a PR when a refactor or removal is worthwhile. All agentic. AI generates code, dashboards, skills, emails at volume. Most teams invest in the generation side and skip the review side. The slop accumulates faster than they can clean it. Full roundtable writeup — including the IKEA effect with skills, "what could you not use AI for this week" as a coaching question, and why token-maxing is the new lines of code: movingavg.com/essays/ai-laye… #AIStrategy #FounderLife #ClaudeCode

Manager roles are about to get pinched. Strategic leadership above, individual contributor work below — and AI is pushing both toward the expert/supervisor zone in the middle. If you manage a design team but haven't shipped a design recently, evaluating AI-generated layouts is going to be hard. Same problem for an engineering manager who can't quickly review AI-written code, or a marketing director who doesn't know which AI ad copy is on-brand. The role you defined two years ago isn't quite the role your company needs now. Two roles will surge in demand instead: quality engineering — quietly out of fashion for years — and data analysis. The signal that keeps a company moving at AI speed without flying off the rails has to come from somewhere. The CEOs running headcount-cut plays are losing their best people while the work still needs doing. The ones investing in supervision and quality are pulling away. Full essay — including who won't make the transition, why "set-it-and-forget-it" agents can't run a sustainable business, and a 4-step start: movingavg.com/essays/ai-wont… #AIStrategy #FounderLife #Leadership

One founder at my AI roundtable hadn't published any AI policy yet, and the reason was specific. He has a team member who reflexively rejects any written rule. Doesn't matter what the rule is. Writing it down triggers a fight. So he wrote nothing. No rule means no friction, but also no protection. The fix isn't to fire the rule-hater or capitulate. It's to skip the handbook and agree on principles together: - We protect our IP. We don't paste sensitive material into free tools. - We use the tools the company pays for. If you need a different one, ask, and we'll buy a license. - We assume AI can be wrong. Important outputs get checked. - We don't put data into tools we haven't reviewed. Most AI rollouts fail because of culture, not technology. A shared understanding of the why is how you ship culture. The handbook comes later — if it comes at all. Full roundtable summary — including why LLMs can't do math, the time-zone bug that gave away the game, and shadow AI hiding inside tools you already pay for: movingavg.com/essays/where-a… #AIStrategy #FounderLife #AIGovernance

Wow. Talked with a corporate executive this morning. They hired internal AI experts to create (vibe code) replacement apps. They've eliminated multiple software companies and are saving millions in tech fees. BUT. They are growing head count as they hire more and more AI experts to create and maintain their home grown tech. Fascinating to see how this plays out!

@kenashley I hope they share what they learn! I think using AI can be incredibly powerful if it supports the bottom line. However, we’ve seen this play out in previous tech revolutions, like mobile apps. Some businesses multiply their revenue. Others simply create a new cost center.