The gap between prompt-driven hacking and production-grade AI-assisted work is engineering fundamentals. Tests, documentation, version control, code review—these matter more now. Running multiple agents in parallel works when you know what you’re orchestrating. Speed exposes who knows their craft.

“Vibe engineering” works because it draws a line. It signals accountability you ship what you’re responsible for, not what AI generates. AI tools don’t level the field; they widen it. Engineers with fundamentals get leverage. The term acknowledges we’re figuring this out while establishing there’s real craft here, not just vibes.

Sometimes the best answer is the simplest one. This short video captures something essential about bias toward action and clarity in decision-making.

The willingness to commit without overthinking. Not recklessness, but confidence in making decisions and adjusting course as needed. Yes as a default stance, not a permanent commitment.

AI coding tools promise 10x speed, but what they actually deliver is 10x more code to understand and fix. The trap isn’t the AI—it’s thinking that faster code generation equals better software.

The real insight here: treat AI agents like lightning-fast junior engineers who need structure, not carte blanche. The solution isn’t less AI, it’s more process around it—specifications, documentation, design reviews, testing frameworks. Raw speed without context just creates expensive technical debt.

What strikes me most is the parallel to tech leadership. Just like keeping complex work to yourself creates team brittleness, relying on AI without building systems around it creates codebase brittleness. Both need intentional investment in structure.

40,000 vibrations per second in a chef’s knife sounds like engineering overkill, but the promise is compelling: cutting with half the effort while reducing food sticking.

I’m curious about durability and whether the vibration mechanism can handle daily kitchen abuse. Cool tech, but I’d want to see real chef reviews before committing to a pre-order.

The distinction between managing and leading becomes crystal clear when you examine how people handle pressure. Sinek’s framework reveals that managers retreat into control mode while leaders step into trust mode.

The most striking insight: information hoarding vs. transparency. Managers treat details like classified intelligence, thinking secrecy equals power. Leaders do the opposite—they call everyone together, explain the challenge, and ask for ideas.

What resonates with me is that Managers dodge discomfort; leaders lean into it with care. “This is tough, but I care about you, so let’s talk.” That’s leadership distilled.

This breaks down how Claude’s memory system works differently from ChatGPT’s approach. Instead of building persistent user profiles, Claude starts fresh every conversation and only retrieves past context when explicitly asked.

The design choice is telling: it prioritizes user control and transparency over convenience. Most AI companies assume users want seamless personalization. Anthropic assumes users want predictable behavior.

Both approaches have merit, but Claude’s feels more aligned with professional workflows where context matters more than continuity. When I’m working on a client project, I want to control what the AI remembers about previous projects, not hope it automatically surfaces the right context.

The broader point here is that we’re still figuring out how AI memory should work. This isn’t the final answer—it’s one experiment in a much larger design space.

Go deep on multiple things early. Don’t choose between specialist vs generalist. Do both. Early expertise compounds into options you can’t build later. Status lags behind reality. Today’s “safe” career paths become tomorrow’s dead ends. Banking, then consulting, then big tech, each generation thinks their safe bet will last forever.

Build global networks online. Best opportunities come through people you meet digitally. Remote work changed the game. Don’t limit yourself to local coffee meetings. Avoid train track thinking. Following predetermined paths (college → grad school → corporate → promotion) feels safe but is actually risky. You’re competing with everyone else on the same track.

The real risk is playing it safe. When you follow the crowd, you compete with the crowd. When you bet on yourself, you compete with almost nobody. Bottom line: The biggest career risks come from playing it too safe, not from betting on yourself.

Performance is something that we will give more time; behavior we won’t. And that’s because behavior is a choice, not a skill set.

We constantly have conversations around prioritization. We can’t do everything, so we have to choose. Not choosing is the worst thing you can do because now you’re compromising everything.

Larger companies … tend to become their own worst enemies. It’s not what the world does to them; it’s what they do to themselves.

Where does ambition come from? It is a lack of adjustment because if you were a perfectly balanced person, you wouldn’t have any ambition. You’re so happy with where things are, you barely have a reason to get up in the morning. But maladjusted people, they just have this disparity between where they are and what they want to do and what they want to prove.

I like people that have attitude, that have a chip on their shoulder, that have a burning need and desire to prove something.

The thing about sales is that great salespeople can’t sell a bad product, but lousy salespeople can sell a great product.

I’d rather hire more slowly but better instead of faster.

“Legacy” is not a word that I use, okay? I find … a lot of self-absorption around that stuff, like I need to live beyond the grave. I really don’t need to, and I don’t want to feel so important that that’s even a question.

Plenty of people speculate about AI replacing coding. Fewer talk about how it works today. Here’s what we’ve seen using Cursor and Claude Code inside a 12-year-old SaaS with about 40 developers.

Cursor works best if you’re fine switching editors. Claude fits better in the terminal and tends to take on bigger, feature-level work. I stick with Claude because I won’t leave Sublime Text.

Productivity is up—maybe 20 percent—but uneven. Agents are great for refactors, chores, and unblocking ideas. They’re weaker at subtle bug fixes and don’t write beautiful code. Over-commenting is common, and you still need human judgment.

The biggest gain is ambition. Solo developers now ship projects that used to take a team. People who once pushed a handful of PRs are shipping hundreds with an agent beside them. Costs are low—our heaviest Claude users barely spend fifty dollars a month.

The hard part of programming is still deciding what the software should do. Turning ideas into syntax just keeps getting faster. In the future, being a great prompter may matter as much as being a great coder.

A very nice this documentary about Fatboy Slim talking about UK DJ culture

The messy, fascinating world of turning lat/lon coordinates into human-friendly addresses. What sounds simple quickly unravels into questions about user experience, data trust, and the real purpose behind OpenBenches.

I find Horowitz’s distinction between good and bad product managers directly applicable to successful CRM architecture - the best CRM architects take ownership like “CEOs of their solution,” anticipate user needs rather than just fighting fires, and focus relentlessly on delivering business value rather than just technical specifications. These same principles apply when designing customer relationship systems that truly drive revenue and adoption. A great CRM architect, like a good product manager, creates clear documentation, makes decisive architectural choices, and measures success through business outcomes rather than feature checklists.

A useful guide on building AI agents based on OpenAI’s recommended approach

I see The Frugal Architect’s laws as essential guidelines for sustainable CRM implementations, where cost consciousness must be woven into requirements from day one rather than treated as an afterthought.

The emphasis on measuring and observing system costs aligns perfectly with modern CRM best practices, where successful architectures implement controls that keep expenses proportional to actual business value delivered.

The recognition that optimization is incremental reminds us that CRM maturity evolves over time, requiring continuous refinement based on genuine usage data rather than untested assumptions about how customers and users will engage with the system.

The Director of one of my all time favorite movies (Dune) picks a few of his favorite movies