“We’re here to put a dent in the universe. Otherwise, why else even be here?” — Steve Jobs

On July 10, 2008, Apple launched the App Store with 500 apps. Investors were skeptical. Developers were cautious. The existing mobile software industry, built on carrier deals and pre-installed bloatware, thought the idea was cute but niche.

Eighteen months later, 3 billion downloads. Over 100,000 apps. Jobs called it “like nothing we’ve ever seen before.” It didn’t just change how software was distributed. It changed what software was. Suddenly, a single developer in a basement could reach 100 million users. The app became the atomic unit of mobile computing.

I think we’re watching the same shift happen right now. Except this time, the atomic unit isn’t an app. It’s a skill.

You don’t need an app for that

Want to track calories? In 2015, you needed someone to build you an app. A developer to write the code, a designer to draw the screens, a backend to store the data, an App Store listing to distribute it. Months of work. Thousands of dollars. And if the app didn’t exist yet, you were out of luck.

In 2026, you tell your AI agent: “Track what I eat today and estimate the calories.” Done. No app. No download. No onboarding flow. The agent understands what you want, figures out how to do it, and does it.

This is the shift. AI agents are remarkably powerful, and they’re getting more powerful every month. They can write code, analyze data, manage files, search the web, and reason through complex problems. For most tasks that used to require a purpose-built application, you can now just ask.

But here’s the thing: a general-purpose agent is like a smart generalist. It can do a lot, but it doesn’t have deep expertise in anything specific. It doesn’t know how Steve Jobs thought about product simplification. It doesn’t know Dieter Rams’ ten principles of good design. It doesn’t know your company’s brand voice or your team’s deployment process.

That’s what skills are for.

What is a skill?

An agent skill is a package of knowledge that transforms a general-purpose AI agent into a domain expert. Think of it as hiring a specialist, except the specialist lives inside your agent and is available the moment you need them.

Anthropic released the Agent Skills open standard in December 2025, roughly a year after their Model Context Protocol changed how AI agents connect to tools and data. Same playbook: build open infrastructure, not proprietary moats. MCP gave agents hands. Skills gave them expertise.

The standard is already supported by 27+ agent platforms: Claude Code, Cursor, GitHub Copilot, Gemini CLI, OpenAI Codex, and many more. There’s a leaderboard tracking over 52,000 skills, with the top ones hitting 180K+ installs. One command to install:

npx skills add openclaw-rocks/skills --skill jobs-ive

Apps gave phones superpowers. Skills give AI agents expertise.

The real bicycle for the mind

I started using Claude Code in early December 2025, building Goban.app. A Go board is a simple thing: a 19x19 grid, black and white stones, a few rules. But something shifted in how I approached the project.

I didn’t need to stress about the technical details anymore. Frameworks, build tools, deployment pipelines: the AI agent handled all of that. And I realized something that changed how I think about building products: technical execution is not what will set you apart. Not now. Not ever again. When everyone has access to the same AI-powered capabilities, the code becomes a commodity. What sets you apart is taste. User obsession. Making it damn simple.

Jobs once called the personal computer “the bicycle for the mind.” He’d seen a study showing that the condor was the most efficient animal in locomotion, with humans ranking somewhere in the middle. But a human on a bicycle beat everything. The computer was that bicycle: a tool that amplified human capability by orders of magnitude.

I think skills are the next bicycle. The computer amplified what you could do. Skills amplify how you think. You don’t just get an agent that codes faster. You get an agent that makes better decisions. That simplifies more ruthlessly. That names things with more precision. That channels decades of accumulated product wisdom at the moment you need it.

It’s always about abstraction layers. Assembly gave way to C. C gave way to Python. Manual servers gave way to cloud. Each layer freed people to think at a higher level. AI agents are the next layer, and they free you to focus on the only thing that actually matters: the person using your product. Skills are how that layer gets its knowledge.

Why Jobs and Ive

Jobs didn’t care about the chip specs. He cared about “1,000 songs in your pocket.” Ive didn’t obsess over manufacturing tolerances for fun. He obsessed because people can sense care, even when they can’t explain it. Their entire philosophy was about resolving complexity so thoroughly that the result feels inevitable.

That philosophy clicked for me while building Goban. When the AI handles the how, you’re free to obsess over the what and the why. And nobody thought harder about the what and the why than Jobs and Ive.

Applying their thinking used to be expensive. You needed a world-class design team, a brand strategist, years of accumulated taste. Now you can encode that thinking into a skill. The expertise becomes something any agent can load on demand, and any person can access for free.

So I built the skill I wished I had.

A Steve Jobs in your pocket

The Jobs/Ive Decision Engine.

Not a surface-level collection of quotes, but a deep, actionable decision framework built from research into their actual methods, principles, and mental models. Seven protocols:

  • Simplify: The elimination test Jobs used to cut Apple from 17 products to 4
  • Name Something: How Apple names products (short, evocative, inevitable)
  • Write Copy: The Apple messaging playbook (“1,000 songs in your pocket,” not “5GB storage”)
  • Design: Ive’s design principles and the Dieter Rams lineage
  • Kill Something: The four-quadrant grid for deciding what lives and dies
  • Price Something: Pricing as positioning, not math
  • Present Something: Jobs’ keynote structure, from “one more thing” to the Rule of Three

Each protocol is backed by deep reference files covering the complete philosophical foundation: real quotes from Isaacson’s biography, Ive’s interviews, Apple keynotes, and the Dieter Rams design principles that inspired everything Apple built.

npx skills add openclaw-rocks/skills --skill jobs-ive

Then ask your agent: “Simplify this landing page” or “Name this product” or “What would Steve do?”

The economics of expertise

Before the iPhone, a professional photo editor cost $699 (Photoshop). After the App Store, you got VSCO for free. The distribution model changed what was economically viable. Skills do the same thing for thinking. A brand consultant charges $300/hour. A skill costs nothing.

And there’s no Apple tax. The App Store takes 30% of every transaction. Skills are distributed through GitHub. No gatekeeper. No platform fee. No 30% cut. The format is open. The only investment is the expertise that goes into writing it. The true bicycle for the mind: Jobs-level product thinking, available to anyone.

This raises an uncomfortable question: can anyone ever charge for a skill?

In theory, you could compile skills into an opaque format and feed them into the LLM through a proxy. But people have always managed to extract system prompts from even the biggest AI providers. Skills are fundamentally the same thing: text instructions injected into a context window. If the model can read them, a sufficiently creative prompt can get them back out.

Pre-configured agents with curated skill sets could be a commercial model. A “product strategy agent” with ten premium skills baked in, sold as a service. But even there, the agent has access to its own configuration. Unless you lock it down completely (which limits its usefulness), it can be asked to share or reproduce what it knows. The protection is only as strong as the guardrails, and guardrails keep getting broken.

I think skills will follow the same path as open-source software. The instructions themselves trend toward free. The value accrues to the people and companies who curate, maintain, combine, and host them. Not to the text in the file, but to the ecosystem around it.

Growing pains

The skills ecosystem is moving fast. Beyond skills.sh and GitHub, third-party marketplaces like ClawHub have emerged as popular hubs, with 3,000+ community-built skills and counting. We published our Jobs/Ive skill there too.

But fast growth comes with real risks. Just days ago, security researchers at Koi Security discovered a coordinated malware campaign they codenamed “ClawHavoc”: 341 malicious skills disguised as legitimate tools, deploying the Atomic Stealer infostealer through fake “Prerequisites” sections. The original malicious skill hit 7,743 downloads before removal. A broader Snyk audit found 7.1% of all skills on the registry contained critical security flaws.

If this sounds familiar, it should. The early App Store had malicious flashlight apps and data-harvesting games. Apple responded with App Review, sandboxing, and entitlements. The skills ecosystem is following the same arc: community flagging (3+ reports auto-hide a skill on ClawHub), open-source scanning tools like mcp-scan, and active security research from Snyk, VirusTotal, and SlowMist.

What’s still missing is a formal pre-publication review process. That will come. Every successful platform eventually builds its immune system. In the meantime: install from publishers you trust, audit unfamiliar skills, and never blindly execute commands from a “Prerequisites” section.

One more thing

The app economy created trillions in value. The skills economy probably won’t, at least not in the same way. I wrote about this in Digital Communism: when intelligence becomes effectively free, everything it can produce follows it down. Skills are text. Text is free to copy, free to distribute, free to remix. The format is open by design. There’s no Apple tax, but there’s also no Apple-sized revenue stream. The economics of skills look less like the App Store and more like open-source software: the artifacts themselves trend toward free, and the value accrues to the people who curate, combine, and care about what they produce.

That’s fine. Maybe that’s the point. The best things in computing have always been free. Linux. Wikipedia. The web itself. If skills follow that path, everyone’s agents get smarter, and the people who care about quality will be the ones who make it matter.

This skill is for the crazy ones. The founders who obsess over button copy at midnight. The designers who delete half the page and then delete half again. The product people who’d rather ship three features done right than ten features done “fine.”

npx skills add openclaw-rocks/skills --skill jobs-ive

Or browse the full skills directory for the 52,000+ skills already available.

This post is also for Lisanne, who always supports me when I disappear into new technology like OpenClaw. She encouraged me to actually post and ship this, instead of keeping it in some private GitHub repo like too many of the things I’ve built before. Thank you for that. Now I have a Steve Jobs in my pocket and Lisanne in my heart. I love you.