Paul
Before I get into this, I want to say upfront that I have nothing against dev shops. Some of them are great. They have real talent, they ship real work, and if you have the resources, they are absolutely the right call. This post isn't about whether they should exist. It's about what I learned when I actually went out and got quotes for a few of my ideas, and what happened when I tried to build the same things myself.
I co-founded and ran a popular website which had millions of monthly visitors called allkpop.com. I did this for nearly 18 years and left in late 2025. After that I had a list of platform ideas I wanted to build. Not blog post ideas, not landing pages, but actual platforms with databases, user accounts, payment processing, the whole stack. Things that would normally require a team and a long time to actually ship.
I have a technical background but I did what most non-technical founders do. I went out and asked for quotes.
The quotes were expensive
I reached out to a few development companies. Some were US-based agencies, some were offshore in Asia and Europe. I gave them all the same brief for one of my platforms. Nothing crazy. A web and mobile app with user accounts, a database, a payment flow, an admin panel, and a few core features that made the thing actually useful.
The quotes came back in the mid six figures. The lowest was around $180,000. The highest was over $400,000. Most of them clustered around the $250,000 to $300,000 range. And those were just the build quotes. None of them included ongoing maintenance, hosting, support, or the inevitable change requests once the thing was actually in front of users.
The timelines were also quite long. The fastest agency said three months for a minimum viable product while most said four to six. One quoted nine months and tried to convince me that was actually fast for the scope.
I sat with these numbers for over a week and the math was not good. Even though I had the cash to spend, I was being asked to risk a third to half a million dollars on an idea I hadn't tested yet. If it worked, it would be wonderful but if it didn't, I'd just torched a year's worth of runway on a single bet. And that was for one platform. I had four or five more ideas I wanted to test.
So I started looking at alternatives.
The cheaper paths weren't necessarily cheaper
The first option was Upwork. You can hire individual developers for roughly 5%-10% of what an agency charges, and on paper that solves the budget problem. But there's more to it than that.
When you hire on Upwork as a non-technical founder, you become the project manager whether you want to or not. You write the specs. You review the code. You manage the timezone gap (usually the savings come from developers in Asia). You catch the bugs the developer didn't catch. You handle the moment three weeks in when the developer goes quiet and you don't know if they're still working on your project or if they've moved on. You're not actually saving money in that scenario. You're trading dollars for hours of your own time, and a lot of those hours go toward managing problems an agency would have solved internally.
The second option was seed funding. Raise some money, hire a small team, build the thing properly. The problem there is that seed funding requires a pitch, and a pitch requires a story, and a story requires either traction or a track record. I had a track record from allkpop, but I didn't have traction on any of the new ideas because I hadn't been able to build them yet. I'd be raising on hope, which means I'd be giving up a lot of equity for a smaller check than I'd want, and I'd be doing it before I even knew if the idea was any good.
The third option was finding an even more technical co-founder. Every solo founder hears this advice. It's bad advice in most cases. The technical co-founders worth having are already building their own things. The ones available are usually available for a reason. And even if you find a great one, you've now turned a solo project into a relationship with all the equity splits, operating agreements, communication overhead, and disagreement potential that comes with it. I didn't want a co-founder. I wanted to build myself.
None of the cheaper paths were actually cheaper once you counted the real costs. Upwork costs you time and code review. Seed funding costs you equity and a pitch cycle. A co-founder costs you half the company and a lot of meetings. The only path that didn't have a hidden cost was the agency one, and that path cost six figures upfront.
Claude Code comes into the picture
I came across Claude Code in the middle of all this and I didn't really expect much. I'd tried ChatGPT and they were fine for basic questions and basic tasks but not for building anything serious. You'd ask these AI LLMs to scaffold something and you'd get back code that almost worked but had subtle bugs, missing pieces, and weird structural choices that you'd spend more time fixing than you'd saved by generating it.
Claude Code was different. After a few hours of using it, I knew I could actually build my platform myself.
I have basic coding skills and some knowledge of devops and infrastructure. I can read code. I can deploy things. I can debug if something obvious is broken. What I can't do is sit down with an empty folder and produce a 50,000+ line codebase while trying to design a polished frontend. That's the gap that used to require a team but Claude Code closed it.
I'd describe what I wanted in plain language. I'd push back when something wasn't right. I'd ask it to refactor when the structure got messy. And I'd watch a working version of my idea come together in real time. Not a prototype. Not a demo. The actual thing, with database migrations and error handling and proper logging.
The first platform I built this way shipped in about a week. End to end. From empty folder to live site on a DigitalOcean droplet with users able to sign up and pay.
Doing the math, DigitalOcean droplets and Hetzner servers run me $20 to $200/mo depending on the size of my platforms. Claude subscription is a $200 a month for the 20X plan. First platform's total build cost was around $300, including domain registration, infrastructure, Stripe setup fees, etc. That's way under 10% of the cheapest agency quote.
What the dev shops are actually selling
I don't think the agencies are scamming anyone. Their costs are real. They have offices, salaries, project managers, QA people, designers, and overhead that has to come from somewhere. After running allkpop I know there's huge overhead with a full-time staff. When they quote you $250,000 for a build, that money is going to actual humans doing actual work. The economics make sense from their side.
What's changed is what that work used to be required for. A lot of what an agency does is translation. They take your fuzzy idea, turn it into specs, turn the specs into tickets, turn the tickets into code, turn the code into a deployed product. Every step in that chain used to require a person, and each person needed to be paid, and the math added up to a six figure invoice.
Claude Code collapses that chain. The translation layer is mostly automated now. I describe the fuzzy idea, the tool turns it into code, I review and adjust, and we ship. The work that used to require five people on a Slack channel now requires me and one terminal. The quality is actually higher in a lot of cases because there are fewer hand-offs where intent gets lost.
The agencies aren't lying about their costs. They're just selling a service whose value proposition is dissolving in real time. A six figure quote for a standard web platform in 2026 is like paying six figures for a custom blog CMS in 2010. The price was fair when it was the only option but it stops being fair when the alternative ships in a week for the cost of an expensive dinner.
Where dev shops still make sense
I want to be careful here because the temptation is to say "agencies are dead" and that's not what I think. Agencies still make sense for a lot of work. If you're a vibe coder without real technical knowledge then agencies will help you a lot.
If you're a non-technical founder who doesn't want to learn anything about code or infrastructure, an agency removes that entire category of problem from your life. You pay the money and the thing gets built. That's a real service and worth real money for the right kind of buyer.
If you're building something with serious compliance requirements, like healthcare or finance, you probably want a team that's seen the regulations before. The risk of getting it wrong is higher than the cost of getting it built right.
If you're inside a large company where the procurement process is easier than learning a new tool, an agency is the path of least resistance. The money isn't yours and the timeline is set by other people anyway.
And if you have a project that requires a specialized skill set, like 3D rendering or low-level performance optimization or hardware integration, that's still where human experts earn their rates.
What's changed is the default. The default used to be "if you want a real platform, hire an agency or a team." That default is wrong now. The new default is "try to build it yourself with Claude Code first, and only escalate to an agency if you hit something Claude Code can't handle." For most standard web platforms, you won't hit that wall.
What I'd tell my pre-Claude Code self
Looking back at the version of me sitting with those six-figure quotes, the thing I got wrong wasn't the math. It was what the math was actually measuring.
The quote isn't the real cost. The real cost is the optionality you lose when you put six figures into a single unproven idea. You don't just lose the money. You lose the next four ideas you would have tested with it, and you lose the speed to pivot when the first one turns out to be slightly wrong.
I also kept telling myself I wasn't technical enough, which was a story based on outdated information. You don't need to be a senior engineer to build a real platform anymore. You need to think clearly about what you want, describe it well, and recognize good code when you see it. Those are skills you can develop in weeks.
And speed compounds in a way the agency model can't match. Shipping a platform in a week instead of four months doesn't just save you the time difference. It changes how many shots you get to take. I've shipped more platforms in the last twelve months than I would have shipped in the next decade paying agencies. Most won't be hits. That's the point. I get to find out, fast and cheap.
The dev shops will adjust. The smart ones already are, repositioning around ongoing operations, and even using Claude Code in their own workflow. The pricing model that built the agency business is the same one that's breaking it. Anyone still quoting like it's 2022 is selling a service whose floor is falling out.
I want to be honest about the other side of this. The same shift that lets me ship a platform in a week for $300 is going to displace a lot of people. The five-person team that used to translate a founder's idea into a working product, that's five jobs. Multiply that across every agency, every contract shop, every offshore dev house, and you're looking at a real wave of people whose work is getting absorbed by tools like the one I'm using right now.
I don't think we should pretend that's not happening. I've benefited directly from it and I'm still a bit uncomfortable with where it leads. The honest answer is that productivity gains this large don't get absorbed by the labor market gracefully. They compound, and the people who lose out tend to be the ones with the least cushion to retrain or pivot.
This is why I think Andrew Yang was early and correct about universal basic income. When the cost of building a software platform drops by 99%, the value created by that drop has to go somewhere. Right now it goes to the founder, the platform, and the model provider. None of it goes to the developers, designers, and project managers whose work used to be the thing being paid for. UBI isn't a perfect answer but it's a serious one, and we need serious answers because this shift isn't slowing down.
I'm going to keep building. But I think anyone using these tools at scale should be honest that the same leverage that makes us faster makes a lot of jobs obsolete, and we owe it to those people to push for policies that actually deal with it.
