I use AI every day at PixelsWithin.
Not because I want to automate everything. I use it to understand ideas more deeply, decide what deserves to exist, and get good products into the world faster.
The MVP isn’t smaller. It’s sharper.
I used to think about a first version as less — fewer features, a trimmed-down core, expand later based on feedback.
That’s not quite it anymore.
The real skill isn’t deciding what to leave out of v1. It’s deciding what to say a hard no to — permanently, not “later” — because it was never actually part of the vision. And then using AI to make the rest of the product as complete as it needs to be, right away.
Say you’re building a video app. The old-world MVP instinct says: ship a basic version, skip the fancy stuff, add hashtag support to the caption box in v2. The way I actually think about it now: if people are just trying to scroll and watch, don’t build them a caption tool with manual hashtag tagging at all — have AI hashtag everything automatically in the background, better than a person would, and never surface that decision to the user as a “feature” they have to think about.
That’s the difference. It’s not doing less. It’s being ruthless about what deserves human attention and what AI should just handle, invisibly, from day one — so what does ship is comprehensive where it matters, not padded where it doesn’t.
Four weeks to first launch is still the target. What changes is what “done” looks like at four weeks.
How we actually get there
The old version of this page described launching, watching what happens, and iterating from there. That’s still true eventually — but it undersold what happens before launch.
Here’s what’s closer to real: I sit with an idea and draw it — with AI — over and over. Not once, not a single spec document. Repeatedly, from different angles, until the idea isn’t fuzzy anymore. Until it’s understood clearly enough that both of us — me and the AI — actually get what we’re building and why.
That’s what makes building the real, whole thing fast instead of risky. It’s not “ship less so we can learn.” It’s “understand deeply first, so building comprehensively is easy instead of a gamble.”
Launch still matters. Once the product is in the world, real customers show us what they use, where they get stuck, and what actually helps their business. Then we keep working on it.
AI vs. slop
I stand by the core point: AI makes it trivially easy to produce a lot of software, and that’s not the same as producing good software.
AI isn’t just a typing assistant speeding up implementation while I do the “real” thinking. It genuinely does a lot of the thinking with me — exploring directions, surfacing problems in an idea before we’ve built it, and doing work that, honestly, impresses me regularly. I don’t want to pretend that’s not exciting, because it is.
What stays mine: deciding what the product is for, what we say no to, and when something’s actually ready. That part hasn’t changed.
What good means
Good doesn’t just mean “not slop.” It means:
- Easy to change and grow — the codebase and product decisions don’t calcify the moment you ship
- Live in the world and making money — not endlessly refined in private. A product that’s brilliant internally and never reaches real customers isn’t good. It’s stuck.
AI in PixelsWithin
When you start a conversation in PixelsWithin, you’ll usually talk to AI first.
You don’t need a finished plan. You don’t even need to know exactly what you want yet.
You can start with the idea as it exists in your head.
AI can help you talk it through, organize it, and figure out what a useful first version might look like.
Then, when there’s something real to decide or commit to, that’s me.
