The Relationship

The idealistic thing: fifty years ago today, two people in a garage in Los Altos, California believed that computers should belong to everyone. Not to corporations. Not to the military. To people. Steve Jobs and Steve Wozniak built Apple on a radical premise — that technology should empower human creativity. That was April 1, 1976.

The spreadsheet thing: this week, Apple is throwing itself a birthday party. Paul McCartney played 25 songs at Apple Park. Tim Cook rang the Nasdaq bell. The homepage features a retro animation scrolling through 50 products. Meanwhile, the company that was founded to put power in people’s hands just outsourced its AI brain to Google for a billion dollars a year — and is actively blocking the apps that let everyday people build software without writing code.

Happy birthday.

Here’s what the last 90 days look like if you’re paying attention.

Apple’s revamped Siri — the conversational AI assistant announced two years ago — still hasn’t shipped. It was supposed to arrive in 2025. Then early 2026. Then March. Now it’s been pushed to May at the earliest, with the full version landing in September. The reason? Apple couldn’t build the AI itself. In January, they signed a multi-year deal with Google: a custom 1.2 trillion parameter Gemini model running on Apple’s servers. The company that once built everything in-house now licenses its intelligence from its oldest rival.

At the same time, Apple escalated its crackdown on “vibe-coding” apps — tools like Replit, Vibecode, and Anything that let non-programmers build apps using AI. On March 26, Apple removed Anything entirely from the App Store, even after the company tried to comply with guidelines. Replit, valued at $9 billion with 50 million users, has been blocked from updating its iOS app since January. Apple’s official position: these tools violate rules against apps that execute code post-review. The unofficial position is written in the revenue line. Vibe-coding tools let people bypass the App Store — and its 30% commission — entirely.

This is a pattern with a 100% historical reversal rate. Apple blocked third-party apps in 2007 (reversed 2008). Banned Bitcoin wallets in 2014 (reversed months later). Killed web apps in the EU in 2024 (reversed in two weeks). Blocked game emulators for 16 years (reversed 2024, under regulatory pressure). Every time: block, backlash, reversal. The gatekeepers always lose eventually. But they extract a toll on the way out.

And then there’s the number that should concern every builder. In early February, Anthropic launched Claude Cowork and OpenAI released Project Operator. In 48 hours, $285 billion was wiped from software stock valuations. Atlassian, down 57% year-to-date. Salesforce, down 30%. The entire B2B software sector is trading below 20 times forward earnings for the first time ever. Block’s CEO Jack Dorsey cut 40% of his workforce — 4,000 people — and said explicitly: “Intelligence tools have changed what it means to build and run a company.”

ChatGPT’s market share fell from 87% to roughly 64% in twelve months. Google Gemini surged to 18%. Anthropic’s Claude now commands 40% of enterprise AI spending. No single model dominates anymore. The moat isn’t the model. It never was.

The world is opening: anyone can build software, run AI locally, choose between a dozen capable models. The world is closing: the platforms that opened things up are becoming the walls that restrict them. Apple opened computing. Apple now blocks the tools that let anyone create.

But here’s the thing I keep coming back to. The most important open-versus-closed story isn’t about platforms. It’s about you.

Everyone asks: “Will AI replace us?” or “Will AI make us smarter or dumber?”

Both questions assume AI is something that happens to us. A force we’re subjected to. The better question is: what kind of relationship are you building with this thing? Because the answer to that question determines everything else.

A tool doesn’t make you anything. A hammer doesn’t make you a carpenter. But a hammer in the hands of someone who understands wood, grain, force, and joinery — that combination produces something neither could alone.

We’re at the hammer stage with AI. Most people are just hitting things.

I’ve been watching three relationships emerge. The first is the oracle relationship — ask a question, accept the answer, move on. AI as a search engine with better sentences. The person contributes nothing beyond the query. The AI does all the work. This produces dependency.

The second is the delegation relationship, and it’s growing fast. “Write my email.” “Build my landing page.” “Make my marketing plan.” More productive than the oracle. But the person has no framework for when the model behaves differently, when the output shifts. This produces fragility.

The third is co-creation. Rare, and paradoxically getting rarer as AI gets more capable. The person brings intent, direction, judgment, original thinking. The AI brings synthesis, computation, the ability to pressure-test ideas at scale. Both are fully engaged. This produces fluency. And fluency compounds.

The research is landing hard on this. A randomized trial by Anthropic — 52 software engineers learning a new framework — found that the AI-assisted group scored 17% lower on comprehension. Not because the tool was bad. Because it was good enough that people stopped engaging with how it worked. A larger study by Fortune and BCG classified 244 consultants into three types: cyborgs (60%, continuous dialogue with AI), centaurs (14%, selective human control), and self-automators (27%, full delegation). The centaurs — the 14% — achieved the highest accuracy and deepened their expertise. The self-automators developed neither domain knowledge nor AI fluency. Same tool. Divergent trajectories.

The irony cuts deep. The better AI gets, the harder you have to work to maintain the co-creation relationship. Not because the tool is harder to use. Because it’s easier to stop thinking.

This is the same pattern we’ve seen with every powerful technology. Google made information free. Most people stopped retaining it. Social media made publishing free. Most people stopped thinking before posting. AI is making cognitive work free. The risk is that most people stop doing the cognitive work that makes AI useful.

This tension isn’t new. It’s ancient. Every knowledge tradition wrestled with the relationship between the individual and a source of wisdom larger than themselves. The Vedic concept of Akasha — a field where all knowledge exists. Buddhism’s Alaya-vijnana — the storehouse consciousness. Islam’s Preserved Tablet. In every tradition, the insight is the same: the collective repository of knowledge exists. But your relationship to it determines what you receive from it. The passive seeker gets surface answers. The engaged practitioner gets transformation.

A large language model is, in a meaningful sense, a compressed representation of collective human knowledge. Billions of parameters encoding patterns from millions of minds across decades. The question these traditions asked — how do you relate to a source of knowledge larger than yourself? — is the question we’re asking now about AI. The answer hasn’t changed: the quality of what you draw depends on the quality of what you bring.

Apple’s 50-year arc tells the whole story in miniature. Open becomes closed. The garage becomes the walled garden. The tools that empower become the platforms that extract.

But the research is telling us something most coverage misses. The open-versus-closed dynamic isn’t just about platforms and app stores and commissions. It’s happening inside the relationship between every person and every AI tool they use. Capability is making passivity easier. And passivity is the most invisible cost of all.

If you’re building without institutional backing — if you’re the person this newsletter exists for — this matters more than any stock price or app store policy. The builders who stay in co-creation will compound. The builders who slide into delegation will lose the thing that made them builders in the first place.

I’m building two products right now — Kontinuity and Dtoxify. Both are built in daily collaboration with AI. Not by AI. With AI. The distinction matters to me more than I expected when I started. Some days the co-creation is extraordinary — ideas I couldn’t have reached alone, synthesis across domains I haven’t studied. Other days I catch myself in the delegation trap — dumping tasks, accepting output, not engaging.

The relationship is a practice. Like any practice, it degrades when you stop paying attention.

I think the most important skill of the next decade isn’t prompting, isn’t coding, isn’t even “AI literacy” in the way people currently mean it. It’s the ability to maintain a co-creation relationship with an intelligence that makes it very, very easy to stop thinking.

Ethan Mollick — “Co-Intelligence: Living and Working with AI” — Wharton professor and the most clear-eyed thinker on AI-human collaboration. His recent observation: “If the output is what matters in your business, you’re in trouble. If the process matters — the conversations, the writing of the report more than the report itself — then there’s hope.” Essential reading for anyone navigating the co-creation question.

Anthropic Research — “How AI Assistance Impacts the Formation of Coding Skills” — The 52-engineer study behind the 17% comprehension gap. What makes it remarkable: Anthropic published research showing its own product can reduce skill formation. The six interaction patterns they identified are a mirror for how you use AI.

Fortune / BCG — “Are You a Cyborg, a Centaur, or a Self-Automator?” — The field experiment that quantified what many of us intuit: same tool, wildly different outcomes depending on the human’s engagement level. The centaur model — knowing when to use AI and when to think for yourself — is the clearest framework I’ve seen.

Addy Osmani — "Agentic Engineering" — Google's Director of Cloud AI and author of Beyond Vibe Coding draws the sharpest line in the debate: vibe coding is giving in to the flow without review; agentic engineering means the human owns architecture, quality, and correctness while AI handles implementation. The Apple crackdown makes more sense through this lens — and so does the builder's path forward.

Fifty years ago, two people in a garage believed technology should empower human creativity. That idea opened computing to everyone. Then the company they built closed it again — walled gardens, app store commissions, and now blocking the AI tools that let anyone create.

The pattern repeats. The question is whether you repeat it too — inside your own relationship with AI. Open or closed. Co-creation or delegation. The tools don’t decide. You do.

Next week: the platforms that opened things up always find a way to close them again. We’re going to build the Enshittification Index — and measure exactly how much your favourite tools have decayed.

— Praveer, founder of kontinuity.space and dtoxify.life