Cursor – Scott Loftesness

I’ve been wondering whether we’ve been looking at the AI stack from the wrong end.

Recently Kris Patel on X laid out a set of excellent questions that he’s looking to have answered as Anthropic and OpenAI move toward going public:

Do you really need frontier-scale intelligence for every task?
Can open-weight models provide an effective alternative to frontier models at a significant discount?
Is the ultimate moat the intelligence or the harness?
What other business models will the frontier labs have to adopt to make the unit economics work long term?
How are you going to prevent distillation from capturing your IP and releasing it?

I’m going to explore only the third question here — moat versus harness. The other four deserve their own consideration, particularly once we have the Anthropic and OpenAI S-1s in hand, revealing for the first time the unit economics of the two largest frontier labs and how much runway they have to support the capacity they’ve contracted.

By “harness” we mean everything surrounding the model: the interface, context, memory, tools, orchestration, evaluation, permissions, and increasingly the user’s accumulated habits and data. The model supplies intelligence. The harness turns intelligence into a product.

Listening to Gavin Baker on the recent All-In episode sharpened this line of thought into something more concrete. He referenced a thought experiment from Eric Vishria: even if OpenAI or Anthropic lost their edge at the pure model layer, they would still retain significant value because of the product harness—the interface, the surrounding tooling, the orchestration—and the user familiarity and habits that have already formed around those platforms. Baker said there is a strong element of truth to it. I think he’s right, and the reasoning behind it is worth spelling out. A model can be replicated, distilled, open-weighted, or commoditized. A mature harness has network effects, switching costs, proprietary context, distribution, workflow integration, and accumulated user behavior. That’s a much harder thing to dislodge.

We are watching intelligence become more abundant and more interchangeable at the same time that the systems built around that intelligence are becoming stickier. On the developer side, the strongest examples are already clear. Cursor turns the IDE into a multi-model agentic environment with deep codebase awareness. Claude Code runs long-horizon coding agents from the terminal, planning, editing, testing, and iterating. Grok Build, Claude Cowork and similar tools emphasize parallel agents and tighter control over local context. In each case the model is a component; the surrounding system does the real work of routing, memory, tool use, and evaluation.

The consumer version of the same idea is now taking clearer shape at Apple. The rebuilt Siri AI shown at WWDC 2026 is not trying to win the pure model race. It is built as a personal harness. A system orchestrator decides what stays on-device with Apple’s Foundation Models, what moves to Private Cloud Compute, and when heavier reasoning is required. Personal context—messages, email, photos, calendar, notes, on-screen awareness—is handled largely on-device through the Spotlight semantic index and App Toolbox. Apple is designing the system so that personal context can be used without giving Apple itself access to it. Conversation history lives in a dedicated Siri app for the user to revisit. And it all syncs across all your Apple devices.

Here is the part I think matters most, and it’s easy to miss if you only read the privacy story. Apple’s Foundation Models framework doesn’t just call Apple’s own models—it’s built to support cloud models from other providers, including Claude and Gemini, conforming to a common protocol. Which means the system orchestrator, not the user, decides which model handles which task. This request goes to the on-device model. That one goes to Private Cloud Compute. A harder one might go to Claude or Gemini. The user doesn’t need to choose, and increasingly doesn’t need to know.

That’s the inversion worth exploring further. The frontier model stops being the interface and becomes a component underneath someone else’s interface. The harness chooses the intelligence. And the company that owns the harness—the OS, the identity layer, the permissions, the apps, the sensors, the notifications, the semantic index tying all of it together—has a form of leverage that has very little to do with whose model is smartest this quarter.

That reframes the subscription question too. I don’t think the right question is whether Siri gets good enough to beat ChatGPT or Claude at reasoning. I think Siri doesn’t need to win that fight at all. It needs to win a different layer entirely—the ambient assistant layer, not the reasoning layer. They’re doing different tasks. ChatGPT or Claude might remain where you go when you think, when I need to reason about something. Siri becomes where I go when I need something done: find (or make) my reservation, text my friend, find that old photograph, update my shopping list, schedule that meeting, add this thought to my notes, figure out when we’re free next week, remind me about that thing we discussed three months ago. Apple’s advantage as an ambient assistant isn’t primarily that it has your personal data. It’s that it has OS-level authority over the world that my personal data lives in.

Of course this is still early. Execution will determine how much of the architectural promise becomes daily reality. Reliability, agentic follow-through, and the quality of the on-device models will matter as much as the privacy story or the multi-model routing. But the strategic bet itself is clear, and it aligns with the broader shift: durable value is migrating toward the systems built around the models, especially systems that sit atop private, permissioned, personal context that competitors cannot easily reach. My early personal experience with the new Siri in iOS 27 betas has impressed me so far. All of this also seems to apply to Google in the context of their Pixel family of devices.

This doesn’t mean frontier labs lose. Pricing power still exists at the high end for the hardest agentic and long-horizon work. Open-weight models will continue to pressure costs and expand access. Distillation remains a real risk. But the more the capability gap narrows, and the more a harness like Apple’s can route among interchangeable frontier models rather than depend on any single one, the stronger the case that value settles into whoever controls the context—not whoever trained the model.

The coming Anthropic and OpenAI S-1s will tell us whether the frontier labs can make their economics of intelligence work. The next generation of Siri, Gemini, ChatGPT, Claude, and whatever comes after them may tell us something even more important: who gets to own the primary relationship with the user.

The model may be the engine. But the harness is where the driver sits.

What a time to be alive!

Did You Really Program That?

The Fundamental Issue

I once found myself in a local restaurant filled with young professors and graduate students from a nearby university. They were clustered around a long table arguing about the nature of originality in a world where machines could now produce human-like text and code with a few keystrokes. I sat at a small table nearby, eavesdropping.

“I just don’t think it’s right,” said a woman with steel-rimmed glasses. “If you’re using AI to write your paper, you should be honest about it. It’s intellectually dishonest otherwise.”

Her companion, a man with unruly hair and a cardigan stretched at the elbows, shook his head vigorously. “But what about the code you’re writing? Aren’t you using GitHub Copilot? Isn’t that the same thing?”

The question hung in the air between them.

The Contested Border

The border between human creativity and machine assistance has always been contested territory. When the word processor replaced the typewriter, did writers suddenly become less authentic? When compilers made it unnecessary to understand assembly language, did programmers become less skilled? Each technological advancement seems to bring with it a fresh anxiety about the dilution of human agency, a sense that we are somehow cheating if we don’t do things the “hard way”.

I recently visited a friend who works at a technology startup in San Francisco. His office was a converted warehouse with exposed brick and polished concrete floors. The ceiling was high enough that you could fly a small drone inside without hitting anything. Software engineers clustered around monitors, wearing noise-canceling headphones and drinking coffee from biodegradable cups. My friend showed me a tool called Cursor, which allows programmers to describe what they want a program to do in plain English, and then generates the code automatically.

“It’s called ‘vibe coding,'” he explained, showing me the interface. “You sort of… gesture at what you want, and the AI figures out how to make it happen.”

I watched as he typed a simple instruction: “Create a function that calculates the Fibonacci sequence up to the nth term.” The AI responded with a dozen lines of code, neatly formatted and commented. My friend nodded approvingly and made a few small adjustments.

“Did you really program that?” I asked.

He laughed. “Define ‘program.’ I told it what I wanted. It wrote the code. I checked it and made a few tweaks. Is that programming? I don’t know. But I’m still responsible for the end result.”

Tools like Cursor and Windsurf are all the rage lately among software engineers as they provide truly dramatic productivity boosts to those writing code.

The Woodworker’s Tools

The discussion reminded me of a conversation years ago with a group of master woodworkers. They were craftsmen who built furniture by hand, using tools that hadn’t changed much in centuries. I asked one of them, a man with fingers gnarled by decades of work, what he thought about power tools.

“People think using hand tools makes you more authentic,” he said, running his palm along the grain of a maple board. “But the old masters would have used power tools if they’d had them. The point isn’t the tool. It’s what you’re trying to create, and whether you understand what you’re doing.”

He showed me a dovetail joint he’d cut with a table saw and jig. “Is this less authentic because I didn’t use a hand saw? The joint is still tight. The wood is still joined. I still had to understand the properties of the wood and how the joint works.”

Writers and programmers alike are wrestling with similar questions. When does technological assistance become a crutch? When does it become cheating? The novelist who uses a thesaurus is not accused of intellectual dishonesty. The programmer who uses a library of pre-written functions is not condemned for laziness. But something about AI assistance feels different to many people.

The Future of Creation?

Perhaps it’s the speed. A process that once took hours now takes seconds. Perhaps it’s the black-box nature of the technology. We cannot see how the AI arrived at its solution, cannot trace the path of its reasoning. We think they’re just dumb machines probabilistically predicting the next word. Or perhaps it’s simply that we are witnessing a fundamental shift in what it means to create.

My programmer friend has a different perspective. “The future of programming isn’t writing code,” he says. “It’s understanding problems and directing machines to solve them. The code is just an implementation detail.”

I wonder if writers will come to feel the same way. Will the future of writing be less about crafting individual sentences and more about directing AI to capture a particular voice or style? Will we come to see the arrangement of words as merely an implementation detail in the larger project of communication? How does this extend to other fields like film, movies and art?

The Disclosure Dilemma

The question of disclosure remains thorny. Should writers and programmers be required to disclose their use of AI assistance? Some argue that it’s essential for transparency and accountability. Others suggest that it’s no different from any other tool, and that the focus should be on the final product, not the process used to create it.

I think of the woodworker showing me his dovetail joint. “The wood doesn’t care how you cut it,” he said. “It only cares that the joint is tight.”

Perhaps the same is true of writing and programming. Many readers won’t care how the words were arranged, only that they resonate. The software user doesn’t care how the code was written, only that it works.

And yet, there is something deep within us that values the human touch, that finds meaning in the knowledge that another person’s mind and hands shaped the thing we’re experiencing. We want to know that somewhere in the process, a human being made choices, experienced frustration and triumph, poured their unique perspective into the creation.

As I left the restaurant I mentioned earlier the debate at the long table was still going strong. I caught a final snippet as I passed by: “It’s not about the tools,” someone was saying. “It’s about the intention.”

Perhaps that’s the heart of it. Not what tools we use, but how we use them, and why. Not whether we use AI, but whether we use it thoughtfully, with intention and understanding. Not whether we disclose its use, but whether we’re honest about our process, both with ourselves and with others.

There’s no question the AI tools are here and that they’re improving dramatically seemingly every day. They’re providing some powerful leverage to amplify our own skills – if we choose to use them wisely.

Note: this initial idea for this post was mine triggered by listening to a podcast interview with Dan Shipper of Every. I had help fleshing it out using Claude 3.7 from Anthropic. The post began with a couple of paragraphs I wrote. Then I used the following prompt: “You’re an expert writer and editor helping me with my personal blog. Write a 1000 word blog post in the style of John McPhee based on the following initial thoughts…” After that I rewrote portions of Claude’s response to add clarity and emphasis before sharing it here.

Note 2: all of this was done on my iPhone.

Share this: