I Built This Entire Product With an AI Partner. Here's What It Actually Cost.

By Brian Crusoe · April 3, 2026 · 8 min read

I need to tell you something about this website you're reading.

Every page. The calculator engine that computes your 3-year TCO. The Stripe integration that processes payments. The lead capture system. The blog you're reading right now. The security headers, the SEO, the cost model behind the numbers — all of it was built by two entities: me, and an AI development partner.

Not a vendor. Not an agency. Not a team of five engineers and a project manager. One person with domain expertise and an AI that writes code, debugs systems, deploys to production, and argues back when my architecture is wrong.

I'm telling you this because I run a site about the true cost of AI implementation. It would be dishonest not to show you mine.

What "Built With AI" Actually Means

Let me be precise about what happened here, because the phrase "built with AI" has been abused to the point of meaninglessness. People use it to mean "I asked ChatGPT to write me a landing page." That's not what this is.

Here's what actually happened:

I provided the domain expertise. I've spent over a decade in manufacturing — MES systems, OEE analysis, TrakSYS, plant floor operations. I know what a $200K vendor quote turns into 18 months later because I've watched it happen. I know which cost categories get buried. I know what plant managers actually need to hear versus what consultants tell them. That knowledge doesn't come from a model. It comes from years of watching AI projects succeed and fail in real facilities.

The AI provided engineering execution. When I said "build a calculator that takes a vendor quote and shows the realistic 3-year TCO with phase-by-phase breakdown," the AI didn't hand me a template. It built the cost model, wrote the FastAPI backend, created the Stripe checkout flow, set up lead capture with SQLite persistence, configured security headers, deployed to a VPS with automatic TLS — and then stress-tested it, found the edge cases, and fixed them.

Neither of us could have done this alone. I don't have 500 hours to write a full-stack web application from scratch. The AI doesn't have 10 years of manufacturing floor experience. The combination produced something that neither could have built independently — and it shipped in weeks, not quarters.

The Real Cost Breakdown

Since this is a site about cost transparency, here's exactly what this product cost to build:

Now let me show you what a vendor would have quoted for the same scope.

I'm not saying those options are wrong. For a large enterprise with compliance requirements and a 50-person stakeholder committee, you need the agency. But for a domain expert with a clear vision shipping a focused product? The math has fundamentally changed.

What the AI Was Good At

The AI excels at things that would have taken me weeks of Stack Overflow and documentation reading:

• Full-stack architecture. FastAPI backend, static frontend, Docker containerization, Caddy reverse proxy with automatic TLS, CI-ready deployment. It knew the right patterns and wired them correctly.
• Security. CSP headers, CORS configuration, rate limiting, input validation, Stripe webhook verification, SSRF protection. Things I know matter but would have taken days to implement correctly by hand.
• Iteration speed. "The calculator should also show build vs. buy vs. hybrid comparison for every input" — and 20 minutes later, it does. With all three approaches computed, formatted, and integrated into the results page.
• Edge cases. What happens when someone enters a $0 vendor quote? What if they select an industry we don't have specific benchmarks for? The AI thought about these before I did.
• SEO and content. Structured data, OpenGraph tags, sitemap, blog post optimization — the mechanical SEO work that's important but tedious.

What the AI Was Bad At

This is the part most "built with AI" stories skip. The AI has real limitations, and pretending otherwise is the same vendor dishonesty I built this site to combat.

• Product judgment. The AI doesn't know that a plant manager evaluating AI for the first time needs the calculator to feel simple, not comprehensive. It would happily add 47 input fields if I let it. Knowing what to leave out is a human skill.
• Domain calibration. The cost multipliers in the calculator — the fact that manufacturing AI runs 3-4.5x vendor quotes while back-office automation runs 2-3x — that's not in any training data. That's from watching real projects. The AI built the engine; I calibrated the numbers.
• Knowing when it's wrong. The AI occasionally generates plausible-sounding cost data that doesn't hold up to scrutiny. Every number on this site was verified against published research (RAND, S&P Global, Gartner, Stanford HAI). Trust but verify isn't optional.
• Strategy. "Should this be a free tool with paid reports, or a SaaS with monthly subscriptions?" The AI can analyze both options, but it can't feel which one is right for this market. I talked to people. I read the room. That's not automatable.
• Taste. The AI's first design drafts were always competent and never distinctive. Every design choice that makes this site feel like it was built by someone who cares — rather than generated — came from human judgment.

What This Means for Your AI Budget

Here's why I'm telling you all of this on a site about AI costs:

The cost structure of building software has changed. Not in the way vendors tell you — they want you to believe their platform is the change and you should pay $200K for access to it. The real change is that domain experts can now ship products that previously required a development team.

This doesn't mean AI is free. It means the cost has shifted:

The last row is critical. AI doesn't eliminate risk — it changes which risks matter. The risk is no longer "can we build it?" The risk is "should we build it, and did we build the right version?" Those are product questions, not engineering questions. And they require domain expertise, not more compute.

The Uncomfortable Implication

If one person with domain expertise and an AI partner can build a functional SaaS product for $300, what does that say about the $200K vendor quotes that land on your desk?

It says the same thing our calculator already tells you: most of what you're paying for isn't the technology. It's the sales team, the account manager, the "customer success" org, the office in a nice zip code, and the profit margin that funds all of it. The actual technology cost — the compute, the models, the infrastructure — is a fraction of the sticker price.

That doesn't mean you should build everything yourself. Complex enterprise deployments with compliance requirements, multi-system integrations, and 10,000 users need professional implementation. But for focused tools, internal dashboards, proof-of-concept projects, and products where one person holds the domain expertise? The vendor model is increasingly hard to justify.

What I'd Tell You Before You Build

1. Your domain expertise is the moat, not the code. The AI can write the code. It cannot replace the 10 years you spent learning why certain approaches fail in your industry. That knowledge is what makes the product valuable.
2. Start with what you know is true. I didn't start with "build me a SaaS platform." I started with "I know vendor quotes cover 20-40% of true costs, and I can prove it with data." The AI built the delivery mechanism for that insight.
3. Budget for judgment, not just execution. The AI work cost me $200. The thinking — what to build, what data to trust, what the user actually needs — took 50+ hours. That ratio is correct. Cheap execution with expensive judgment is the right model.
4. Ship something ugly, then fix it. The first version of this site was rough. It worked. People used it. Then we made it better. AI makes iteration so cheap that perfectionism is the real waste.
5. Verify everything. The AI will confidently generate a cost multiplier that's plausible but wrong. Every number on this site traces back to a published source. This is non-negotiable.