We Are Not Alone
The entire software industry is repricing in real time. So is my life, the output I produce, and what that output is worth. Cost is going to zero, and we all are kind of cooked.
I wasn’t ready for it. I know your feeds are flooded with “AI can now do this” and “I built this as a vibecoder” slop, alongside engineers rushing to the comments to rage or flex. Both sides are annoying but neither is wrong. My situation is just genuinely weird enough that I need to put it out there.
Four years ago I basically couldn’t write code. I was technical enough to fumble through Stack Overflow or some random YT tutorials, but I could not build software. I was a product and growth guy (the annoying kind) who had ideas and no way to ship them unless I had a team and budget, which luckily I mostly did.
Last Friday I sat down at 9 PM to prove a point to a friend. He was demoing something his team built. I started going off about how much the build process had changed and I guess sort of decided to just show him what I could build today. The idea I had in mind was coming from another friend who had a real problem, or maybe not exactly a problem, but something that could be sped up massively.
Context: my buddy Kadir runs Spektra, easily the #1 mobile racing game studio out there. His team was deep in A/B testing some game icons. I’d been messing with recent image-gen models for this exact workflow and getting really good results, but the process had too many steps. Everything was disconnected.
I couldn’t sleep and I really wanted to build something easier to use and navigate.
Sixteen hours later, aso.kitchen was live. Not just a waitlist page or a PRD this time. Production software:
- Next.js with full routing and SSR
- Auth and session management
- Stripe payments with subscription tiers and webhook handling
- Database, image storage, rate limiting
- A landing page that actually converts
Idea is simple. Paste any App Store link, describe what you want, get production-ready icons in seconds. Real output that real developers can submit to the App Store today as the results are pretty good.
I mean it isn’t a revolutionary product. It solves a specific, narrow problem, or speeds up a process that would take longer. Gives an output that actual artist can actually iterate quickly. But that’s the whole point. The interesting part is the delta between what this would’ve cost 48 months ago versus what it cost last Friday (my lack of sleep).
The old math was much simpler: a frontend engineer, a backend engineer, a designer, maybe DevOps, maybe a PM. One to six people, four to twelve weeks, $10K-$200K fully loaded.
That cost structure just collapsed into one AIDHD guy and a Friday night with too much caffeine.
The output isn’t trash as well. Stripe webhooks handle edge cases properly, no leaks, decent rate limiting at multiple levels. Two SWEs I trust reviewed the codebase and both said it’s clean enough to keep building on.
These aren’t the things you wave away in a demo, or used to. These are details that used to take engineers real time to do properly. I don’t think the cost of building software isn’t decreasing, but it might be evaporating. I am not saying it will hit zero, or SWEs are not needed. They are needed more than ever for now nonetheless if you want to scale anything into SaaS level. But… For now.
I should probably be more specific about what the workflow actually looked like because the abstractions matter more than the vibes.
I was still the idea guy, but with more control and needed to do a lot of research along the way. My PC though… Basically a room full of competent junior to mid level (Opus 4.6) engineers operating at whatever speed I could think. We made decisions together about data models, auth strategies, payment flows, API design. We researched popular open source repos, indie dev approaches on X, Reddit threads. I chose Drizzle over Prisma. JWT over database sessions. Upstash over self hosted Redis. R2 over S3.
Every single one of those decisions required product judgment and technical taste that no model (sort of) provides on its own without deep research first.
Then I built each piece iteratively. Generate the schema, review it, adjust relations. Generate API routes, catch edge cases in webhook handling, fix a race condition in credit deduction, wire up the frontend, iterate on UX. Every commit checked in parallel with unit tests and Codex running code review.
Don’t get me wrong.AI doesn’t remove the need for judgment. It removes the tax on executing judgment. Back in the days, my bottleneck was never “can I think of the right solution” It was “I need time and money to find out if I’m right.”
That bottleneck is pretty much gone.
Every mid to large team has a middle layer. People whose entire job is translating between the person who knows what to build and the act of building it. That layer exists because of friction. Friction between having an idea and testing it. Friction of leaking technical knowledge to understand what your idea exactly would translate into. But in today’s reality, the build cost goes to zero, friction goes to zero. And I feel when friction goes to zero, the middle layer has no reason to exist, at least in a large scale.
Which is what we unfortunately did, just this year. All of the middle roles are now eliminated. Our internal production team works as a hacker house on steroids. Each developer owns a project end-to-end, in sort of a role we call “Fullstack Builder”. No designers. It’s hard to justify when you have frontend-skills, Shadcn MCP, Figma MCP (and much more) plus taste. No project/product managers. We create the PRDs together, ideate together, use Linear to piece everything together so there are no sylos. We test out ideas rapidly. Kill it if it doesn’t perform. Double down and pull in more resources (then, middle managers that we can access) if it works to scale it up.
Unfortunately thepeople who hold value are on the edges. Deep domain expertise on one side, knowing what to build, and having visual taste. Deep technical taste on the other, knowing how it should work. The translators in between? Their margin is compressing to zero. Including myself.
If everyone can build software at some point in the horizon, what’s defensible? Not code. It was already commoditizing. AI probably fast forwarded a decade.
Definitely not ideas. Ideas are always cheap. Execution was what always mattered. Now execution is cheap too.
Is it about the moat? Is it about speed? Is it on UA? What’s the key?
I guess what’s truly left is the stuff that compounds. Distribution, data. brand? Is knowledge expertise that important? Maybe it is, but for how long?
I built aso.kitchen because I understand ASO from years of working in mobile growth. I know what devs need at exact specs. I know game studios swap icons for increasing conversion rates and need brand consistency with holiday flair. I know the difference between a decent icon and a great one moves conversion rates 20–30%. But…
I would love to say that none of that came from AI and it came from doing the work and experience. But I’d be wrong. I ran deep researches accross 3 models that got me up to speed with most recent data, comparing various studios/startups and their success at ASO. Plus, AI just let me encode its knowledge + my existing knowledge (which again, I don’t know how much it matters) into a product in 16 hours instead of 16 weeks.
I guess the real moat isn’t “I can build.” People are surprised at what I can do today, but soon enough, everyone will be able to build. I guess the moat is (at least for the time being) “I know something specific about the world that most people don’t or see a real problem, and I can ship a product around that knowledge faster than anyone.”
I don’t think raising a seed round without an MVP will work anymore unless you have an amazing track record. I think in today’s world, most of the times, MVP should already exist. The round should fund distribution, data acquisition, and the first hire who isn’t you or your co-founder.
If you’re at a big company watching this happen, the playbook is simple: smaller teams, higher output, deploy the savings into distribution and data moats. The companies that figure this out first eat the ones that don’t.
Everything downstream of that assumption needs to be re-examined.
I am having AIDHD + AI depression as we speak. I don’t know what’s next. But I know for a fact that nothing will ever be the same.