All plays › Online and software businesses

Online and software businessesHermes agent

Every recurring expense audited against the books: $425,724 in annual cuts, about $30,000 a month net

Picked and checked by Olga Pearson, host of The Practical AI Show · Published

Result

self-reported

From about $1.12M in operating expenses over the prior 12 months, he cut $425,724 a year: $250,572 in contractors, $131,160 in services and $43,992 in software. New AI costs are $5,482 a month, so he puts net savings at $29,995 a month, plus $48,575 in one time builds he says pay back in under 2 months. He also says shipping velocity went up 10x.

Michael Houck runs Rye Valley. He listed everything the company does, how long it takes, how often it happens and whether it is rule based or needs judgment, then had his Hermes agent go through the books to see what they spent on each. Almost 40% of spending could in theory be covered by a company brain (every meeting transcript, Slack message, email and decision in one place) with one agent on top. He cut contractors, services and software, and says it's never fun to say goodbye to people on a team. Some savings went into a client portal, internal tools and new sites.

Who did it

Michael HouckRye ValleyMichael Houck's essay · Oct 15, 2026

Try it now

  1. List every task your company does, how long it takes, how often, and whether it's rule based or needs judgment.
  2. Decide which jobs you want people to keep doing no matter what.
  3. Have an AI agent match your books against the list to see what each task costs you.
  4. Cut or replace the rule based ones in stages, starting with software and services.
  5. Track what the new AI setup costs each month against what you cut.
Copy-paste prompt
Here is a list of everything our company does, with time, frequency and whether it needs judgment, plus our last 12 months of expenses. Match the spending to each task and show me which rule based tasks cost us the most each year.

Make it run every week

A play works once. Turn it into a routine in three small steps:

  1. Run it once and check the result yourself.
  2. Give your AI a schedule. Tell it when to run this play (every week, or every month) and to message you with what it found before you act.
  3. Keep a log. One line per run: what was checked, anything flagged, what you did.

See how owners turn plays into systems →

What to watch out for

Self-reported in his own essay, which promotes his workshop. Most of the savings came from contractors and people on the team, so plan any change carefully and fairly. Safer version: start with software and services before people. Build it with Page: one client portal page so clients and your agent work from the same place.

Did it work?

PageMotor keeps this site free.