AI Agents
Posts filed under AI Agents.
Topic
-
AI prices fell up to 22x. Our bill fell 2x. Here is where the rest went.
At Berry, we use cheaper models to give agents harder work, with a monthly model bill less than half its old size. But making the output useful can take better instructions, not more tokens.
-
You Can Afford More Agents. You Can't Afford to Watch Them.
I have spare tokens but not spare attention. I capped the tasks I follow at once, but I stopped counting dropped threads afterward, so I cannot claim the cap improved the work.
-
Cheap Mistakes Go to AI. Expensive Ones Wait.
Routine AI work becomes dangerous when an error passes every control and sits unnoticed. Sort tasks by the thinking they need and the consequence of escape, then measure the checking layer with sampled escape rate and time to detection.
-
Your AI Employee Has a Rate Card. You Set It.
I paid twenty times too much for the same agent work because I picked the wrong model. The bill needs an owner, and the checking work belongs in it too.
-
Document the Process First, or the Agent Scales the Chaos
Our first agent impressed everybody and changed little because the job still lived in managers' heads. Useful work started when we wrote that knowledge down.
-
What Is an AI Employee?
An AI employee is a role your firm has to define and support. The hard part is checking the work before it consumes the senior time you meant to free.
-
Fewer Managers, More Managing
I've watched supervisors trade hours of report preparation for minutes checking AI output. The benefit depends on using that time to develop the team.
-
The AI Policy Template That Fits on One Page
Give your team a clear way to use AI, with approved tools, a real stop before client-facing actions and someone responsible for the checks.
-
Delegate the Inputs, Own the Outputs
A task being easy to check doesn't mean anyone checks it. Give agents room to prepare work, with a named owner for the checks and the client send.
-
Your AI Is Just a Better Google Search
If AI can't reach your company information, you keep carrying the work into and out of a chat. Connect one workflow safely, write its rules and check whether the whole job improves.