Agents with a human in the loop
Designing work where AI executes and a person approves — and knowing when not to use an agent.
Agents with a human in the loop
- Series
- 04
- Class
- 01–03
- Take
- —
- Product
- Sidekick+
- Host
- Andrés
- Guests
- Edixon
- 04·01 What an agent is (and isn't) To record
- 04·02 The approval point To record
- 04·03 When the agent promises what it can't deliver To record
Sidekick+ installs a workforce: senior people plus AI. AI operates, the human approves — nothing that matters goes out without a person approving it.
We started out selling “an AI crew” and changed it in October. The series tells why, and teaches how to design that balance: which task you give an agent, where the approval point goes, and how you notice an agent promising what it can’t deliver.
The other series
The framework for working with AI
How a company decides, prioritizes, builds and measures its work with AI — without buying tools before understanding the problem.
Software for property management
How a trade that lives in paper, group chats and spreadsheets — running a residential building — becomes a web and mobile app.
An AI SaaS, done right
Building a vertical SaaS with AI from zero to production: spec, stack, tests and deploy — and what's still missing.
An agent that keeps your books
You tell it what happened and the agent records it, tags it, simulates scenarios and flags patterns. You decide.
An AI seller that gets the meeting
Automated prospecting with judgment: signals, messages and reading replies — all the way to a meeting a person closes.