This past week I watched a guy on Twitter (X) build a free replacement for a product people pay $12 a month for, mostly by himself, for fun.
His name is Andrew. He calls it his "war on paid software that should be free." He picks a target, spends a few days on it, and ships. First it was Granola, the AI meeting notetaker. He built Quill, an open source version that runs entirely on your Mac, records both sides of a call, transcribes it on device, and costs nothing. A day later he picked Wispr Flow, the voice dictation tool, and shipped Parrot: hold a key, talk, watch the words appear at your cursor. Also free. Also open source. Also, by his own account, a few days of work.
Almost nobody has heard of either project yet. This isn't a funded startup with a launch plan. It's one person with an opinion and an AI coding agent.
Two weeks earlier, at the opposite end of the spectrum, Jack Dorsey's Block open sourced Buzz, a free replacement for Slack and for the code-hosting tools engineering teams live in every day, built so AI agents and human employees work side by side in the same conversations, with the same permissions, on equal footing. Whatever AI agent you run, Claude Code, Codex, Block's own Goose, it just plugs in. Within days, tens of thousands of developers had already picked it up and started building on it.
Three launches, two weeks. A public company on one end, a guy most people have never heard of on the other. Same instinct pointing in the same direction: the tools that make software cheap to build are also making software cheap to replace.
The feature was never the point
Here's the part that actually made me sit up, though, and it has nothing to do with how fast either project caught on.
Granola wasn't the first AI meeting notetaker. It wasn't even close to first. The first wave just dumped your transcript into a generic summary format and called it done, and people tried them and mostly stopped using them, because a summary that doesn't look like how you actually take notes isn't useful, it's just another document to skim. Granola's real move was learning your note-taking style during the call and shaping the output around it, then asking for feedback and getting better at matching you specifically. The outcome people were buying was never "AI writes my notes." It was "I get to be fully present in this conversation and still walk away with notes I'd have taken myself." That's a completely different thing to build, and it's why Granola won a category that had already been "solved" twice before.
Which means Quill replacing the transcription feature doesn't automatically replace the outcome. It might. I don't know yet. But it's worth being precise about what's actually being copied.
Same thing with voice. A few months ago Peter Steinberger, the guy behind OpenClaw, was on every podcast talking about running four to ten coding agents in parallel, entirely by voice, because his hands couldn't keep up with what he had to say anymore. At the time I remember thinking that sounded like a power-user party trick. Then I got deeper into Claude Code myself, started having more to direct and less patience for typing it, and picked up a tool called Willow to talk to my agents instead. It's been a real shift. The outcome isn't "voice-to-text." It's "the rate I can hand off ideas finally matches the rate I actually think of them."
What this means for the moat
I picked up a framing a while back, from Alex Hormozi's writing on offers, that's stuck with me: people don't buy features, they buy outcomes. It sounds obvious until you actually audit what you're selling.
Anyone can build a website builder. I know, because I did, in a week, with Claude Code. Anyone can build a newsletter tool, a booking page, a CRM. If what I'm selling is "Coachstack has these features," I'm exactly as exposed as Granola and Wispr Flow are right now. The moat was never going to be the feature list. It's the outcome underneath it, translated for one specific person: a coach who wants a business that runs itself well enough to leave the day job for, built from a hundred small decisions about how coaches actually work, not from a spec sheet. That's much harder to vibe-code in a weekend than a single feature, because it isn't one thing. It's the whole shape of the product, tuned to a specific outcome for a specific person.
I don't think any of this is bad news, either for me or for software in general. Open source getting this good, this fast, is mostly a good thing. Andrew isn't taking anything from anyone. He's proving the tools now exist for anyone to build what they need instead of renting it. Some of Granola's and Wispr Flow's users will probably move. Most probably won't, because paying for something that just works, and letting someone else own the infrastructure headache, is worth real money to a lot of people. Maybe the honest read is that AI just grew the whole pie: more builders, more options, more people who'd never have shipped software a year ago shipping it now, for fun.
But if you're building something people pay for, this is a good week to ask whether they're paying for the feature or the outcome underneath it.
The question I'm sitting with, and maybe you should too: if someone vibe-coded your product's features this weekend, would they have actually replaced what people are paying you for?
See you next Saturday.
Peter
