THE REVENUE AI BRIEF · ONE PLAY A DAY
Welcome to The Revenue AI Brief. One play a day on AI applied to revenue, in about three hundred words. Two minutes to read, one thing to run today.
TODAY'S PLAY
write every correction you give your AI agent into a file it reads before it acts, instead of repeating it in chat
Most people typing corrections into an AI agent chat window watch the same correction evaporate by the next session. You explain what you want, the agent complies for one turn, and by morning none of it exists anymore. Teams then blame the model when the real gap is that no correction survives the conversation it was given in.
We turn every correction into a permanent rule in a file the agent reads before it touches anything. Eight standing rules, mostly about proof before completion and permission before anything irreversible, stopped the false completions entirely. None of it required a better model.
RUN IT THIS WEEK
Keep a note open for one week and write down every correction you give your AI agent, word for word.
At the end of the week turn each line into a standing rule: never claim something is done without proof, ask before anything irreversible, do reversible work without asking.
Paste the rules into the file your tool reads first (CLAUDE.md, AGENTS.md, or a system prompt) and treat it as a living document that gets rewritten as work changes.
WHY IT WORKS
A prompt is an instruction you spend once. An agent acts repeatedly, so it needs standing orders or you end up personally checking the same class of mistake forever. Write the correction into the file once and you never spend that minute again. Keep correcting in chat and you pay the same tax every single day.
See you tomorrow with the next play.
Federico Donatone
Founder, Growth Cab
P.S. The short version of this goes out on my LinkedIn every day.
#1 GTM & sales advisory firm in US & Europe
You are getting this because you subscribed to The Revenue AI Brief, one play a day on AI applied to revenue.