What Used to Take a Team
For most of my career, shipping something real meant hiring. Designers, engineers, a roadmap measured in quarters. I ran that machine at Intridea and again at every company after. The ideas were rarely the scarce part. The scarce part was the people and the months it took to turn knowing a domain into working software.
That math changed.
What AI did, said precisely, is collapse the middle between expertise and a shippable product. I already knew advertising, media, video, how invitations fail, how SMS behaves, how CTV platforms work. I already knew how to architect software. What I didn’t have, as one person, was the hours to turn that into several deep products at once. Now I mostly do. The model writes the code, drafts the UI, generates the artwork, stands up the boring infrastructure. I point it at a problem, throw most of it out, and ship the part that holds up.
Taste is still the job. The model is fast, not tasteful.
Since May I’ve been building a lot. Not because I set a calendar challenge. Because the cost of trying dropped, and when that cost drops, the thing that matters is which bets you make with the knowledge you already have. Software background gets you to a coherent architecture. Domain knowledge tells you which product is worth the weeks. AI fills in the middle that used to require a team.
The products so far:
- Gatherd, party invitations that arrive as text messages and handle RSVPs over SMS. Live.
- Splatchat, video chat with AI characters. Live.
- Wavemaker, AI video generation from a URL, topic, or prompt. Live.
- Waverunner, multichannel performance advertising from a single URL. Live.
- Waveband, self-serve DSP for agencies and brand teams: seven channels, one seat. Live.
- Vidiyo, run your own 24/7 streaming TV channel, free, on every major platform. In private beta.
- Waveform, a site builder that works backward from how every other one does it.
None of these are demos or landing pages with a waitlist. They’re usable, and most of them are live. One person builds each one. I write up what I made and how, including where the AI workflow helped and where it fought me.
The point of listing them isn’t the volume. It’s the shape of the work. Each one sits on something I already understood. Gatherd on how people actually RSVP. Wavemaker, Waverunner, and Waveband on years of ad creative and performance. Vidiyo on FAST and CTV. Splatchat on what video chat wants once characters can talk back. The depth comes from the domain, not from the tooling. The tooling just stopped being the reason I needed six people.
Running alongside this is a separate set of AI experiments: The Hartwells, Wrong Version, thescrollhole, Imaginary Ventures. Those aren’t products. They’re me poking at what AI can sustain and make you believe. Different question, different track.
I don’t know how long this window stays open in this form. Models get better, tools get better, and so does everyone else’s access to the same stack. For now, the scarce input is still judgment about what to build, not the ability to build it.