Work · YouTube · @salstack.ai
SalStack
Tutorials on building agents that hold up
Most AI-agent tutorials give you boxes, arrows and buzzwords. This one shows the complete process.
Short, complete walkthroughs on building AI agents for real work: define the job, choose the model and tools, design context and memory, build the loop, test where it fails, add evaluations and security, prepare for production.
The problem
Most “AI marketing agents” are ChatGPT with extra confidence. They invent claims, repeat the same idea and take actions without controls. Most tutorials teach the enormous prompt, which is not an agent, it is an unpredictable draft generator.
The approach
Every tutorial ships the full operating contract: five source-of-truth files, a repeatable campaign brief, independent scoring against a rubric, one human approval before distribution, CPL- and ROAS-based optimization, permission boundaries and a kill switch. The model reasons; software enforces the rules; the human makes one decision.
What is true today
Three tutorials published in the week of September 8, 2026, including “Stop Building Fake AI Marketing Agents” and “Build Your First ChatGPT Marketing Agent (One Approval, Full Feedback Loop).”
The question I ask
What part of your marketing workflow would you automate first?
Related notes
Written along the way.
Most “AI marketing agents” are just ChatGPT with extra confidence
They can invent claims, repeat the same idea and take actions without proper controls. Here is how to build a safer one.
Read ↗One enormous prompt is not an agent. It is an unpredictable draft generator.
A beginner tutorial on turning ChatGPT into a controlled marketing system with one human decision: approve, revise or reject.
Read ↗From basic concepts to advanced agent architecture, in six minutes
Most AI-agent tutorials give you boxes, arrows and buzzwords. This one shows the complete process.
Read ↗