A studio of one, with ten AI agents.
I'm one person, not a software company. So I built a small one out of AI agents: each has a job, a checklist and a folder of shared notes. They draft, test and report. I decide.
Ten roles, like a real studio
The same jobs a big studio has, from the first idea to the fixes a year after launch. The agents run on Claude.
DIRDirector
Reads every agent's report each morning, keeps the plan, and brings me the few decisions that are actually mine.
RESResearch
Scores new app ideas against real demand, watches trends and competitors, and kills weak ideas early.
DESDesign
Writes the product spec, the screens and the store graphics, and keeps each app's voice consistent.
ENGEngineering
Writes and fixes code on branches, upgrades dependencies, and prepares builds.
QAQuality
Tests builds, runs the release checklist, and watches crash and freeze rates after launch.
LEGLegal & privacy
Audits what an app collects (the answer should be nothing), drafts policies and store declarations.
ASOStore listing
Writes and tests Play Store titles, descriptions and screenshots.
GROGrowth
Plans and drafts posts, and checks every claim against a list of facts before I see it.
SUPSupport
Reads reviews and email, drafts replies, and turns complaints into bug reports.
DATData
Reads the Play Console numbers each week and tells me what changed and why.
Anything public, paid or permanent waits for me
Every task has a colour. The colour decides whether an agent can just do it.
Green
The agent does it and logs it: research, reports, drafts, code on a branch, internal test builds.
Amber
The agent prepares it completely and waits for my approval: posts, review replies, emails, store listing changes.
Red
Only I do it: signing keys, prices, payments, legal filings, and publishing anything to the public store.
A few things are never allowed at all: adding ads or trackers, buying reviews, or making a claim about an app that isn't on its list of checked facts.
What it looks like day to day
| When | What happens |
|---|---|
| Every morning | Support reads new reviews and email; the director writes me a short brief with anything that needs a decision. |
| Monday | Data reads the week's numbers from Play Console. |
| Thursday | Growth drafts next week's posts, checked against the facts list, for my approval. |
| Sunday | Research looks for new ideas and trends and scores them. |
| Once a month | Engineering, quality and legal run maintenance: upgrades, a security check, the paperwork. |
Why tell you this
Because you should know how the things you install are made. AI does a lot of the drafting here, and I think that's worth saying out loud rather than hiding. What it doesn't do is decide what goes out under my name. That part is still me, and it's the part I'm answerable for.
So far this setup has produced two apps, Bloom and Life OS, and the pages you're reading. If you want to see how it holds up, become a tester.