A live marketplace, rebuilt from scratch.AI agents build it on their own, for days at a time.

DailyPB is the pickleball app I built and run. Players find games, sign up and pay. Hosts get paid through Stripe. It outgrew the no-code tool it started on, so I'm moving the whole thing into real code.

A team of AI agents does the building. I set the plan, make every call, and run the method that proves each piece works. You can watch every job, live.

Watch it live I can do this for you

Built in publicEvery job on a live board, from plan to proof.

Not a demo. A live business with real money.

Most AI builds you see are a to-do app or a landing page. This one has thousands of players, hosts who get paid through it, and a history of games and payments that can't get lost.

Everything has to move, and nobody should notice.

Real money

Payments and host payouts through Stripe. Same accounts, so no host signs up again.

About 5,300 accounts

The old tool can't hand over passwords, so every player gets back in with Google or an emailed login link.

About 86,000 sessions

Every past game, signup and payment comes along, checked count by count.

Repeating games

Weekly sessions across time zones and daylight saving, edited the way Google Calendar does it.

A brand-new database

64+ tables, with privacy rules on every one. The browser never touches the data directly.

Ready for AI assistants

Players can ask ChatGPT or Claude to find a game and sign them up. Built in from the start.

AI writes code fast. The hard part is knowing it's right.

Fast code nobody checked is how a business gets hurt. So every feature goes through the same steps, and none of them depend on a person catching mistakes.

Read the old app

Every page, workflow and rule, turned into files an agent can read

Decide

Keep, cut or change, written down

Define done

In plain words, before any code

Tests first

Built from real data, and they fail first

Build

One slice at a time, data moved with it

Attack

A different AI model tries to break it

Prove

Every test passes, twice, then a real browser run

Write it down

For builders, for users and for AI agents

Agents run the orange steps. Deciding is mine.

Every feature is written up three ways, before it counts as done.

For the next builder

How the feature works, how to drive it, and its traps. Any agent can pick it up cold.

For players and hosts

The help center page is updated with new screenshots. Then every other help page is searched for the old way and fixed.

For AI agents

The tool names, descriptions and API spec an outside AI needs. The test: a fresh AI agent, given only those docs, has to sign up for a game on its own.

It runs alone for a day or two at a time.

No one at the keyboard. A boss agent hands each whole job to a sub-agent. The sub-agent runs the full method on its own and reports back when it's done. Then the boss hands out the next job, around and around.

24 to 36 hoursHow long one boss session runs alone.

~500,000 tokensThe boss's memory limit. When it gets there, it writes down where things stand and a fresh boss takes over.

The boss agentHands out whole jobs

One whole job each, many at once

Sub-agent

Runs the whole method for one job, on its own

BuildReviewFixProve

Sub-agent

Runs the whole method for one job, on its own

BuildReviewFixProve

Sub-agent

Runs the whole method for one job, on its own

BuildReviewFixProve

↺ Done: reports back

Built like money depends on it. Because it does.

These are the checks that run without me. They're why I can trust work I didn't type.

No double-bookingA race test fires signups at the exact same moment on every action that changes shared data. Two people never get the last spot.
Money adds upA ledger that is only ever added to. A refund is a new line, never an edit. A daily check matches Stripe, the ledger and host payouts to the cent.
Same answers as beforeOld app and new app, side by side. Real cases from the old app run through the new one, and the answers have to match.
Tests you can trustEach test is broken on purpose to prove it really catches the bug it was written for.
No surprise at the endThe data moves with every slice, not in one scary weekend. If old records don't fit the new design, the slice isn't done.

Plus a list of automatic checks before every merge: security scans, privacy tests, backup restores and more. Plain code, not AI, so no AI can talk its way past them. See the full list ›

Agents build. I make the decisions.

My job is the plan and the calls only an owner can make: what to keep, what to cut, what to change, and what "done" means.

The build never stops to wait for me. If it hits a question the plan didn't answer, it takes the safe path, writes the question down for our next planning session, and keeps going.

That's how I'd run an AI project in your business, too.

A real screen from the new app, with made-up sessions and real DailyPB court photos.

"Done" is a scorecard. Written before launch.

Not my gut feeling and not an AI's opinion. The pass marks were set from the old app's real numbers, and the switch is judged against them.

Data

Every count and total matches. 200 random players' history checked.

Money

$0.00 difference between Stripe and the books, every day for 30 days.

Behavior

100% of the old workflows mapped to new code, a test, or a deliberate cut.

Speed

Pages load in under 1 second for 3 of every 4 real phone visits.

Logins

Zero players locked out a week after the switch.

Security

Every "who can see what" test and an outside attack scan pass.

Don't take my word for it. Watch it.

Every job is on a public board: what it's for, what "done" means, what the review found and how it was fixed.

See the live build board

Outgrown your software? Let's talk.

A 30-minute call. Tell me what's slowing your business down. I'll tell you what AI can build, and how we'd know it works.

Book a call

Or email chuck@startuptosold.com