Writing

What running our own agents actually taught us

Every piece here comes out of a real incident on our own machines, with the dates and commits attached. If a rule only works on our setup, it does not get written up.

The install tax: your agents are paying for skills they never use

I cut a 748,980-token instruction dump to about 5,500. The session still carried 84 skill descriptions bidding to fire. Includes the two-question sort you can run on your own box in ten minutes.

Read it

The feasibility check: how to know if an app idea is worth building before you build it

I asked my phone for a tree identifier app while standing in a forest. It ran a feasibility check before writing a line: can the hard part work, and should this exist at all. Four steps, a cap, and a verdict that is allowed to be no.

Read it

Claim drift: your agents are still saying the thing you already corrected

I deleted a false claim from my website on July 8. On July 27 I found my own agent still drafting from it, in a config section labelled safe to state as fact. Includes the ten-minute audit for finding every copy a claim has left behind.

Read it

AI app builders without a subscription: how one-time licenses work

The three ways AI coding tools charge, the cost maths behind bring-your-own-key, and the six questions worth asking before paying once instead of monthly.

Read it

Build on your own box

freeapp.ai is a voice-controlled coding agent that runs on your own Linux machine, on your own model key. 300 founder spots at 299 USD, one time, with lifetime updates.

See the founder license