How we work · October 19, 2026

We Vibe Code Every Client Product. Here Is What Tutorials Skip.

We build every client product with AI. The first 90% takes a few prompts. The last 10% takes most of the time. What lives there, and how to budget for it.

Let us start with a confession that surprises people. Every product we build for clients is vibe coded. All of it. We do not hand-write code and then sprinkle AI on top. We describe, the AI builds, we review, we correct, and we ship.

So this is not a post telling you vibe coding is dangerous or that you should leave it to professionals. It is the opposite. Vibe coding is the most important change in how software gets made in a generation, and we have bet our company on it.

It is a post about the one thing the tutorials leave out. It is the thing that decides whether your idea becomes a product or stays a demo.

The first 90% is real

The videos are not lying. Open Lovable, Bolt or Claude Code, describe a client portal for your firm, and within a few prompts you will have something that looks like a finished product. Login screen, dashboard, forms, a chat assistant that answers questions. It looks like 90% of the job.

In a sense it is. It is 90% of what the product will eventually contain. That part of the build has become almost effortless, and anyone who has watched it happen for the first time knows the feeling. It is genuinely astonishing.

The last 10% is where the time goes

Here is the part that matters. Ninety percent of the time it takes to build a finished product goes into the last ten percent of it. That was true before AI and it is still true now. AI compressed the first part dramatically. It did not compress the second part nearly as much.

That is the time and resource commitment most founders do not see up front. They see a product that looks nearly done after a weekend, and they reasonably assume the rest will take another weekend or two. Then months go by.

Google's Addy Osmani describes the same gap from the engineering side. We see it from the builder's side on every project. The number changes depending on who is counting. The shape does not.

What actually lives in the last 10%

When we say "the last 10%," we do not mean polish. We mean the work that makes a product safe to put in front of paying clients. A few examples from the kind of products we build.

Every path a real user can take. Your prototype works because you know how to use it. Real clients skip steps, go back, refresh halfway through, use an old phone, paste text from a PDF that contains strange characters, and open the same page in two tabs. Each of those is a small thing. There are hundreds of them.

Permissions that are actually right. A client should see their files and nobody else's. An account manager should see their clients and not the whole book. An admin should be able to fix a mistake without a developer. Getting this right is not one prompt. It is a design decision for every screen, and then testing every one of them.

The AI behaving well when it matters. A demo assistant only has to answer the questions you thought of. A live one has to handle the question you did not expect, refuse the one it should not answer, and never confidently tell your client something wrong under your brand.

Money, email and the things around the app. Payments that confirm on the server, emails that land in inboxes instead of spam, a real domain, backups that have actually been restored once, secret keys stored somewhere safe. Our guide to where a vibe-coded app lives goes through each piece.

Changes without breakage. By the time the product is live, every new request touches something that already works. Without discipline, each change risks breaking something else. This is the part where non-technical builders most often get stuck in a loop.

Feedback from real users. The biggest item of all. The first time real clients use the product, they will tell you, mostly through their behavior, that some of what you built is wrong. Correcting it is part of the last 10%, and it is the part that turns a tool into something people pay for.

Why the tutorials skip it

It is not a conspiracy. The last 10% just does not make good content. A video of someone building an app in an hour is exciting. A video of someone testing permission rules for the fortieth screen is not. So the content you find when you search is almost entirely about the first 90%, and the impression it leaves is that the first 90% is the job.

For a hobby project, that impression is harmless. For a service business putting a product in front of its own clients, it is expensive. You plan for a weekend and commit to a quarter.

How to budget for it

If you are planning to build a client-facing product, whether you do it yourself or with help, here is how we would think about the commitment.

Assume the demo is the start, not the middle. When you have something that looks nearly done, you are at the beginning of the real work. Plan your time and your expectations from there.

Budget the last 10% as the larger share. If the first version took a week, do not plan for one more week. Plan for the remaining work being several times longer than what you have done so far.

Decide who owns it before you launch. A live product needs someone who answers when it breaks and who keeps improving it. If that is you, block the time. If it is someone else, decide who now.

Ship to a small group first. A handful of friendly clients will find most of the last 10% for you. Far cheaper than having all of them find it at once.

Keep the prototype and the product separate in your head. The prototype proves the idea and captures the decisions. The product is what survives contact with real users. Sometimes the second grows out of the first. Sometimes it is cleaner to rebuild once you know what you are building.

The skill that matters now

If AI handles the first 90%, what is left for people? Two things. Deciding what is worth building in the first place, and carrying it through the last 10% to something clients rely on. Neither is about typing code. Both are about judgment.

That is where we spend our time. We vibe code the product, the same way you could. Most of our effort goes into the parts the tutorials do not show: deciding what the product should be, and getting it through the last stretch.

How we work with this

We build you a working prototype before any money changes hands. It is the first 90%, and it is a real thing you can click through, not a slide. That lets you decide whether the idea is worth the last 10% before committing to it.

If it is, there is one fixed build fee for getting it to a finished product. After launch we stay on as product lead, because the last 10% never quite ends, and we are paid as a share of what the product earns rather than a flat retainer. If you have an idea and want to see your first 90% before deciding anything, book a call.