How I turned the operations of a SaaS company fully AI-run.

We bought BetterPic when it made $1.5K a month.

We had run an agency, SaaSmic, for years and wanted a product of our own. AI headshots were early. The market had 3 or 4 players.

We scaled it to $275K a month in spring 2025, and three sources did most of that work. Medium articles brought $287K in 2025 from 2 or 3 posts. Google search brought $264K. One Reddit thread brought $175K. Affiliates added another $40K to $50K in the best months, close to 20% of revenue.

The team grew with it, and so did the ambitions. A CFO, a CMO, a product manager, 3 developers, 2 affiliate managers, support in shifts, and a second team of 7 building BetterStudio, an AI fashion-shoot product. About 25 people at the peak. We tried Facebook ads and display along the way. They never paid back.

Then the big models shipped image generation good enough for a photoshoot.

People opened ChatGPT and made their own. The 3 or 4 competitors became 30. Revenue went to $106K in January 2026 and kept sliding to $40K by May. Since then it has grown about 2% a month, to $43K in August.

A one-time product with no urgency behind it keeps one channel through that: demand capture. Be there when someone searches. That is the whole marketing plan now, and it is why the routines below look the way they do.

One expensive lesson from that year. I was CMO then, and I only knew each channel as I discovered it. Affiliates I understood through SEO and the listicles, and not much deeper.

So I hired 2 part-time affiliate managers with different skill sets, together for the price of one full hire. Three reasons: learn from them, see if they could open corners of the channel I could not, and find out if they could beat me at that game.

After 6 months they had never added more than 5% on top of what the listicles already did. That money bought me the ceiling of the channel, and the rules I wrote from it run every morning now.

Layoffs are the part people skip in pivot stories. We did not get to skip them.

At the peak the group had about 25 people on payroll: BetterPic's own team, a second team of 7 building BetterStudio, and designers and AI engineers shared across the products. We went to 5 people. Two of them are developers who work full time on Patricia and come to BetterPic on call when an order is stuck.

The pivot was simple to say and slow to do. In March we took a week away from the desk and made one decision: BetterPic pays for the next venture, so BetterPic has to run itself. Every recurring job in the company went through one filter.

Does this need hours, or does it need tokens?

We started with Claude Code in March. Patricia moved into our Slack in August, and the routines moved with her.

This is what runs on its own today, by department.

Numbers first, opinions second. These are the six I repeat on every call with a founder.

Ricardo Ghekiere, my co-founder at Patricia, uses this ladder with agencies. Most companies he meets are on level 2 and believe they are done. BetterPic sits on level 5: the routines run on their own, and my hour a day is judgment, taste and the approvals.

If I had to rebuild all of this in another company, I'd start with one department: search.

It's where BetterPic's money came from. Medium, Google and one Reddit thread made $726K in 2025. It's also the most repetitive work in the company: reports, page fixes, briefs, outreach, Reddit replies, and now the question of what ChatGPT says about you.

So search is the loop Ricardo and I install for other teams first. Eight routines per channel, one a day and three a month. Every routine ends in a draft a person approves.

You can. I did. Here is what it cost.

Six months of work, from the pivot in March to today. Two part-time affiliate managers for 6 months, to learn rules that now fit in one file. And this summer my Google Ads budget scaler died silently for 51 nights before I noticed.

Today a watchdog checks the routines, and every routine names its sources. All of it is simple to write and slow to get right. The install is those months, skipped.