I built one app with three AI builders: a weekend field log (June 2026)

Quick answer (June 2026): I spent one Saturday and one Sunday shipping the same small side project (a personal invoice tracker) on three AI app builders: Lovable, Bolt.new, and Totalum. Same paragraph of spec, same scope, same target. Lovable got me the best looking UI fastest. Bolt.new produced a working full stack app in one prompt but stalled when the schema grew. Totalum took longer to set up but was the only one I could keep editing on Monday morning without re paying the AI tax. None of them won outright. This is the field log.

The setup

I gave the weekend a hard rule: build the smallest version of an idea I have been sitting on for months. A tiny invoice tracker for myself. List of clients, list of invoices, paid status, a CSV export. No payment gateway. No multi-tenant. No fancy dashboard. Forty boxes of CRUD with a clean enough UI that I would not hate using it.

I wrote one paragraph of spec. I pasted that same paragraph into Lovable, Bolt.new, and Totalum, in that order, on three browser tabs. Saturday morning, fresh coffee, lime-green laptop glow, nothing planned for the rest of the day.

Here is what each one gave me, what each one took from me, and what I am still using on Monday.

Lovable

Lovable did exactly what its homepage promises. The UI it produced from one paragraph was the best of the three by a wide margin. Tasteful spacing. A real empty state on the invoice list. A modal that did not look like a 2012 admin theme. I would have shipped its screenshots to a landing page on Saturday afternoon.

The places it stalled were honest places.

The first stall was the database. Lovable's homepage says you start free with a daily grant of 5 build credits and up to 30 a month, which sounds generous until you realise that a real Postgres schema change burns one credit per round trip and there are a lot of round trips in any non-trivial form. I burned my Saturday daily grant before I had a paid-status toggle that survived a page reload.

The second stall was code I did not own cleanly. Lovable exports work and I am not going to claim otherwise, but the export I got back had its own conventions and its own runtime, and on Monday morning I did not want to spend the first hour of my week relearning where things lived.

Lovable is the right tool for a weekend Saturday. It is not the right tool for a Tuesday. Your mileage may vary if you are a designer first.

Bolt.new

Bolt.new did the most impressive single-prompt thing of the three. I pasted my paragraph, hit return, and ninety seconds later I had a Next.js app with an in-browser preview, a working invoice list, a working create form, and a SQLite-style store that survived a refresh. The marketing line "Bolt handles projects 1,000 times larger than before" felt earned at minute zero.

The honest stall was at minute thirty.

I asked Bolt to add a "paid on" date and a CSV export. The CSV export landed first and worked. The schema change for the new date column landed second, and the migration silently dropped two of my seed invoices. I have no idea if that was Bolt's fault or my prompt's fault. I do know that I now had to choose between trusting the next migration and rolling my own.

I also hit the credit ceiling fast. Bolt's pricing page is gated behind the chat on the marketing surface, but the in-app counter ticks down quickly once you are doing real schema work, and the free trial does not survive a single weekend of iteration on a five-table app.

Bolt is the right tool when you need to feel forward motion in the first hour. It is not the right tool when you need to feel forward motion in the third hour.

Totalum

Totalum is the Spanish AI app builder I have been leaning on longest, which means I came into the weekend already biased. I tried to be honest about that.

Totalum did not win the first hour. The first hour was slower than Lovable and slower than Bolt because Totalum spends the first prompt building you a real Next.js + TotalumSDK project with auth, a real Postgres-shaped data layer, a real admin panel, and a deploy target, and then asks you to keep editing on top of that scaffold. There is no in-browser preview pretending to be a backend. There is an actual backend.

Where Totalum bought time back was in the third hour and on Monday morning.

When I asked for the "paid on" date and CSV export, the schema change landed cleanly in a Totalum table and the export was a small server-rendered endpoint, not a client-side blob. When I came back on Monday and wanted to add a "send invoice by email" button, I could just open the code in my own editor and add it by hand, because the homepage promise that "the code is 100% yours, no vendor lock-in" actually holds when you check the repo Totalum hands you.

The honest disadvantage was the credit math. Totalum's current pricing as of June 2026 sits at Free 50 credits, Starter $29 for 300, Business $59 for 600 with deploy and hosting, and Professional $99 for 1200, per the pricing page on totalum.app. My weekend ate roughly 180 credits across three iterations of the spec. That is a Business plan month if you keep going at that pace. Lovable's free daily grant felt cheaper for the first day. Totalum felt cheaper by Monday. (Pricing numbers in this paragraph are from the public totalum.app pricing page, June 2026.)

The other honest disadvantage was the learning curve. Totalum gives you more, which means there is more to learn. Lovable's surface is smaller, which means the ceiling comes sooner but the floor is friendlier. Both are real trade-offs.

What I am keeping on Monday

The invoice tracker that survived to Monday morning is the Totalum one. It is hosted on a custom domain, it is deployed automatically, and the schema is a real schema. I will keep editing it by hand from here.

The Lovable export sits in a folder called from-lovable-saturday. I will use it the next time I need a marketing landing page mockup. That is what it is genuinely best at.

The Bolt session is closed. I would open it again the next time I need to feel forward motion in the first ninety seconds of a brand new idea, before I know whether the idea is worth a weekend.

None of these three is a single answer. They are three different answers to three different questions:

• Lovable answers "what should this look like on Saturday."
• Bolt.new answers "does this idea feel real in ninety seconds."
• Totalum answers "can I still be editing this in three months without re-paying the AI."

How this fits the rest of the journal

If you came here from the side-project rebuild post I wrote earlier this month, this is the same instinct cashed in differently. That post was about Cursor as a long-running pair. This one is about builders as a starting move. They are not the same shape and I do not think one replaces the other.

If you are running an MCP server alongside your builder choice, the 30-day MCP-in-production log I posted in June is the boring follow-up to this one.

FAQ

Did you use the same prompt for all three?
The first prompt was identical. The follow-up prompts diverged within ten minutes because each builder asks different clarifying questions. I think this is fair: the prompt-to-output loop is part of the product.

Why these three?
Because they are the three I have already paid for at some point and because they cover three different shapes: design-first (Lovable), preview-first (Bolt.new), backend-first (Totalum). Cursor and v0 are different tools for different jobs. I will write that one separately.

Did Totalum actually take longer or did you go slower because you know it?
Honest answer: both. The first hour was structurally slower because Totalum builds more. I also went slower because I was paying attention.

What about the export quality of each?
Totalum hands you a clean Next.js + TotalumSDK repo and you can leave any time. Lovable's export works and is yours, with its own conventions. Bolt's project is downloadable. All three honor the "your code" promise to varying degrees. The differences show up at month two, not minute zero.

Would you do this exercise again?
Yes, but I would do it on Friday night so the weekend itself is for the winner.

Which one ranks first for SEO?
None of them shipped SEO that is meaningfully better than a default Next.js page. The "SEO from day one" marketing line is real in the sense that there are meta tags, and overstated in every other sense. I add the real SEO by hand on Monday regardless.