A marketing strategist I'd been introduced to that week stopped me ten minutes into our first call. "Can I ask you something direct?" Sure. "Is this a feature, or is it a business?"

The Question I'd Already Answered

He wasn't being rude. It was the sharpest question anyone had put to me in months, and I told him so — I appreciated the directness. Most founders flinch at that question. I didn't, and not because I'm fearless. I'd spent the previous several months living the answer.

Over those months I had taken my own software spending apart line by line and watched it fall from about $4,000 a month to roughly $750. The difference wasn't frugality — I had rebuilt the tools myself. In the same stretch I went from zero commits on 📝GitHub to nearly 4,000, almost all of them produced alongside an 📝AI agent that hadn't existed in usable form a year earlier. I would not have called myself a software engineer when the year began.

That exercise handed me a kind of x-ray vision. When you rebuild a dozen products with your own hands in a single season, you stop seeing "apps" and start seeing two very different things. There were the tools I could reconstruct in a weekend — a scheduler, a transcript cleaner, a dashboard. Stripped of their branding, each was a thin layer of logic sitting on top of someone else's 📝API, and once I had seen that, I couldn't justify paying rent on it. Then there was a smaller set of things I could not reconstruct at all, no matter how many agents I aimed at them.

By the time the stranger asked his question, I had already sorted my entire stack into those two piles. Everything I'd deleted without hesitation had been a feature. The one thing I had built and could not walk away from — could not have rebuilt even if I'd wanted to — was something else entirely. His question wasn't a threat. It was a sorting test, and I had already run it on myself.

The Graveyard, and Why It's Filling Faster

The distinction isn't mine, and it isn't new. Software has always kept a graveyard for products that turned out to be features.

📝Apple has filled that graveyard so reliably the industry named the practice after one of its victims. In 2002, a search tool called Watson was made pointless overnight when Apple shipped the same capability inside its own Sherlock app. Developers started calling it getting "Sherlocked," and the list has grown every year since: Konfabulator absorbed by Dashboard, Growl by Notification Center, the webcam app Camo by Continuity Camera, Tile by AirTags, the journaling app 📝Day One by Apple's own Journal, 📝Grammarly's core function by the writing tools now baked into the operating system. A product is a feature when the platform can ship it in a release and erase you in a keynote.

Artificial intelligence took that dynamic and set it to fast-forward. The boom from 2023 through 2025 produced tens of thousands of companies that were, honestly, a prompt behind a login screen. When 📝OpenAI shipped its GPT Store and its enterprise tier, an entire category — "fine-tune a model on your data" — stopped being a business in a single announcement. Andrew Chen of 📝Andreessen Horowitz wrote about the coming "revenge of the GPT wrappers" in a world where the models themselves are commoditized. The test investors now apply is blunt: if you could rebuild it in a weekend hackathon, it is a feature. Even OpenAI, the platform doing much of the absorbing, now lectures the next generation of founders about defensibility. @r

And the cost of doing the absorbing has collapsed. 📝Boris Cherny, who created 📝Claude Code at 📝Anthropic, has described some of its own engineers burning hundreds of thousands of dollars in model tokens every month — more compute, monthly, than a senior engineer is paid in salary in a year. Sit with what that means competitively. A company can hand a single engineer more in monthly tokens than that engineer earns in a year, and turn them loose. If your product is a feature, the threat is no longer that a platform clones you on its roadmap. It can clone you in an overnight multi-agent run. That is why the strategist's question has become the question — the one in every investor meeting and every founder's quiet hours. A decade ago, "feature or business" was a risk to weigh. Now it is the first thing anyone serious wants to know.

What Can't Be Cloned

So the question has a real answer, and panic is not it. If a platform can clone any feature in a night, survival depends on building the things that are not features — the things that do not yield to compute.

📝Vidya Narayanan framed that answer more cleanly than I have seen anywhere else: the 📝trifecta of moats is data, user experience, and community. Three layers — and they are not equal. Data is proprietary information that improves the more it is used; a rival cannot clone the years of accumulated usage that produced it. User experience is the felt quality of the product, the reason a person reaches for it by reflex — and it is the weakest of the three, because a well-funded team can study good UX and reproduce it. Community is the hardest to build and the hardest to kill: when people stay for each other rather than for the software, there is nothing left for a platform to ship. You cannot release someone else's relationships in a keynote.

That is what the trifecta resolves. A feature is one layer, almost always the copyable one. A business is the three of them compounding. Anthropic can ship any capability 📝MythOS has; it cannot ship a person their own library, and it cannot ship them the people they have come to think alongside. MythOS — the thing from my audit I could not rebuild — was built deliberately on all three: data in each person's memos, experience in the editor and the knowledge graph, community in the libraries people share and fork into one another's. The technology underneath is simple. That was always the point.

So when someone asks whether you are building a feature or a business, don't reach for a defense. Reach for an inventory. Name the data that compounds. Name the experience worth a reflex. Name the community that would grieve if you vanished. Name all three, and you have a business. Name only the feature, and the platform already has your release date — it has just not told you yet.