When Your Agent Stops Being Agentic: Now What?
⚙️ Re-engineering agents into non-agents to scale your organization.
What happens when an agent consistently selects the same action because what it knows and observes has become 100% predictable? Is it still an agent when, for all intents and purposes, it behaves like a traditional piece of software? Why not replace it with traditional software?
Remember KISS? Keep It Simple, Stupid. It’s an ideal we’ve worked towards in leadership, management, and design for longer than we’ve been alive. We shouldn’t stop now. We should be telling our agents to kiss their keisters goodbye because their final form might not be agents at all, but something deterministic, predictable, and stable.
Let’s travel back to the 1980s and look at something complicatedly simple. Pac-Man.
Are those four ghosts chasing Pac-Man using AI? It certainly looks that way. But they’re actually using traditional IF/THEN programming logic to determine how to engage Pac-Man, with decisions evaluated as a ghost moves through the maze. And they don’t all use the same decision logic. Each has a distinct targeting approach.
- Blinky (Red): Pursues Pac-Man directly.
- Pinky (Pink): Targets a position ahead of Pac-Man.
- Inky (Cyan): Uses a more complicated targeting calculation.
- Clyde (Orange): Switches between pursuit and retreat depending on how far away he is from the yellow guy.
Nice, simple rules.
Get Chris Pehura’s stories in your inbox
Join Medium for free to get updates from this writer.
You can predict how those ghosts will move based on Pac-Man’s position, their own positions, and the game’s rules. That’s not necessarily the case with complex agentic systems. You can’t reliably predict the outcome simply by knowing how you would move Pac-Man.
So, use agentic AI when you don’t sufficiently understand the environment or can’t reasonably predict the outcomes. Stop using it when the environment and outcomes become predictable.
At that point, replace your agentic AI with something built around the kind of simple, deterministic rules we see in Pac-Man. And there’s absolutely nothing wrong with that. But why stop there? Why not have our agentic AI create and deploy its own replacement, then sunset gracefully by sending your CAIO a retirement speech by email?
💡 Pac-Man is the metaphorical gold standard for agentic system replacement.
⚠️ Make boring sexy again.
The challenge with AI is that it’s exciting. It involves more emotional investment, presents more interesting problems, and can even offer a stronger sense of job satisfaction.
Just making something work is boring.
So make boring sexy again.
❗ Non-AI solutions become apparent when your agents have mapped the environment, cracked the code, and solved the puzzle. Replacement solutions become so obvious that even an AI can build them.
You might as well let them do it.
