A founder's note · six months in

I used to be the software.

Before AI could write code, I ran online stores the normal way: a team of assistants, a folder of procedures, and my own attention holding it all together. This is the story of moving that whole system into software.

THE PROCEDURES THE PROGRAM while store: sense() decide() act() EVERY PROCEDURE A PERSON CAN WRITE DOWN, CODE CAN RUN
01

How it used to work

Every store I ran before this one worked the same way. Virtual assistants in different time zones. A knowledge bank full of procedures: how to answer a customer, when to turn off an ad, when to reorder stock, how and when to pay a supplier.

The procedures were good. That was never the problem. The problem was that every single one of them needed a person awake, paying attention, following the steps. Training took weeks. People left and took the training with them. And the hardest calls always came back to one person anyway: me. I was the fallback for everything. I used to be the software.

And it worked. These are my own screenshots from that life: single days of sales from the stores my team and I ran by hand.

Store dashboard screenshot: 1,095 euros of sales in one day Store dashboard screenshot: 2,208 euros of sales in one day Store dashboard screenshot: 2,557 euros of sales in one day Store dashboard screenshot: 3,476 euros of sales in one day Store dashboard screenshot: 4,813 euros of sales in one day Store dashboard screenshot: 7,395 euros of sales in one day Store dashboard screenshot: 10,084 euros of sales in one day Store dashboard screenshot: 15,460 euros of sales in one day

The receipts. Different stores, different days, from the years before DLFT. Every euro here was earned the old way: people following procedures, and me holding the whole thing together.

02

The turn

Then AI learned to write software, and one specific thing changed: any procedure you can write down, you can now turn into code that runs itself. Not a prototype. A working program, built in days instead of months.

So we stopped hiring for the procedures and started compiling them. Three engineers, six months, one store. We took the knowledge bank apart and moved it into code, procedure by procedure. The AI wrote the software. The software runs the store.

03

The handover

The old procedures, next to what they became. These are real, and the dates are from our own project history.

The procedureCheck the ads every morning. Turn off the ones losing money.
NowA program checks every ad on a schedule and turns off losers by itself. running since 6 Apr
The procedureWhen a customer asks where their order is, look up the tracking and reply politely.
NowSoftware reads every customer email and answers the routine ones with real tracking info. Hard cases still go to a person. fully autonomous since 19 Jun
The procedureWatch what ads the competition is running. Save the good ones.
NowA scanner studies competitor ads around the clock. Hundreds of thousands collected and analyzed. running since 13 Mar
The procedureMake new ads when the old ones tire. Give budget to what sells.
NowThe software drafts ads, launches them, and moves budget to the winners on its own. since 7 Apr
The procedureReorder stock before it runs out. Track every parcel.
NowStock is watched continuously, orders go to suppliers automatically, and every shipment is followed to the door. since 30 Mar
The procedurePay the supplier once the order ships. Top up the ad account when it runs low.
NowPayments run themselves, with a guard against ever paying twice. since 22 May
authored by mehrag.faiz · human · intent: tell it like it happened
04

Where the experiment stands

The hypothesis was easy to say and expensive to test: can an entire online business run on autopilot? Six months in, the store sells, ships, advertises and pays its bills with no staff. Real customers, real money, real parcels.

You do not have to take my word for it. The store publishes its numbers, live, straight from its own records.

Watch the store run itself
05

What comes next

Right now the AI writes the code, and the code makes the decisions. That order matters: code is predictable, testable, and cheap to run a thousand times a day.

The next step of the experiment is letting AI make some decisions directly. It will earn that the same way the code did: measured against humans, on a short leash, until the numbers say it is ready. That is how our customer service went autonomous, and it is how everything else will.

First the knowledge left my head and became a team. Then it left the team and became code. The experiment is what it becomes next.

Running a brand the old way, like I did? Talk to me.

Write to the founder