You still start with an idea. Something annoys you, or you see a gap, or you just want a thing to exist that doesn’t. That part hasn’t changed.
What has changed is how quickly the idea runs into reality. It’s less “can we build this?” and more “why would anyone care, and how would they even find it?” Models and copilots can spit out code and copy that looks fine on the first pass. The bottleneck moved. It’s less about typing speed and more about taste, structure, and knowing what not to ship.
Discovery is weird now too. Traditional search still exists, but it’s not the only front door it used to be. People get pulled in through feeds, recommendations inside tools, AI summaries, word of mouth, or whatever their stack already ships with. So you end up shaping the idea twice: once for the product, and once for the world it has to live in. Not in a sleazy way. Just honestly. If nobody is going to stumble on you through a five word Google query, you need another path: integrations, APIs, community, partnerships, or a problem so sharp people go looking for the category even when they don’t know your name yet.
That can feel cynical if you let it. Like you’re not building what you believe in, you’re building what fits. I don’t think it has to be that bleak. It’s more like editing. You still have a point of view. You’re just forced to say it in fewer moves.
Then there’s the crowded room. Do we even need new software? There is already a lot of everything. And yeah, sometimes the answer is no. Another thin wrapper on top of a model is not a company, it’s a weekend. The stuff that still feels worth doing is usually narrower: a specific pain big players ignore, data and permissions that need to stay sane when generation is cheap, or a workflow that has to stay reliable even when the fun parts are automated.
I swing between optimistic and skeptical about all of this, often in the same afternoon. On good days, cheap generation is freeing. You spend time on the fuzzy human stuff: who this is for, what “good enough” looks like, what breaks if you’re wrong. On bad days, it feels like everything is a remix and you’re shouting into a hallway full of people selling the same hallway.
If you’re a founder, a buyer, or someone who just ships things, the questions end up similar even if the vocabulary changes. Who is actually stuck on a random Tuesday? What would they use if you vanished? What would make them switch anyway? If you can’t answer those without squinting, the problem isn’t AI. It’s clarity.
So creating software in this era is a bit stranger, a bit faster, and a lot more honest about distribution. You’re not the only person who can code anymore. You might still be one of the few people stubborn enough to hold a standard when the tools say “done.”
Top comments (0)