← Back to blogYouTube Video
Published June 17, 2026
Build deterministic protections into your workflows
Can't play the video or having issues? Here's the direct link.
Summary
Without deterministic protections, what you have isn't an AI agent — it's a ticking time bomb that goes off the moment the LLM hallucinates. I define deterministic protections as plain code or scripts you write yourself that stop an agent from taking drastic or malicious actions, and I walk through the simplest one to implement: the AI never makes permanent changes to your database autonomously. A human reviews and approves those writes, or they don't happen. My position is blunt — if those guards aren't built in, the thing isn't a workflow or an agent at all.
Key takeaways
- Never let an AI make permanent changes to your database without manual human review and approval.
- Deterministic protections are code you write — scripts that sit between the model and any drastic or malicious action, blocking it outright rather than asking the model nicely.
- If an agent has no deterministic protection mechanisms built in, treat it as unreliable by definition, not as an autonomous agent.
Description
If your AI agent doesn't have deterministic protection mechanisms built into it, it is not an autonomous agent. It is a ticking time bomb and completely unreliable.
Transcript
Build deterministic protections into your AI workflows. If you don't have deterministic protections built into your AI workflows and agents, what you have is not a workflow or an agent but a ticking time bomb that's going to explode in your face sooner or later when your LLM inevitably hallucinates.
You need to implement deterministic protections into your AI workflows. Now, what do I mean by deterministic protections? Well, this is basically just some code or script or some sort of deterministic program that you write that's going to prevent the AI agent from taking any drastic or malicious actions.
For example, a super simple example, super easy to implement: don't let the AI make permanent changes to your database, at least not without manual user review and approval. Unless a human being is reviewing and approving it, don't let the AI modify your database autonomously.
Dude, you’re at the bottom of my landing page.
Just book a call already if you’re that interested.
You scrolled all the way here.