Hi Reader,

A lawyer in Australia recently thought they'd cracked the code on AI safety.

They used Claude AI to research case law. Then—because one AI wasn't enough—they "validated" everything with Microsoft Copilot. Belt and braces. Double-checking. The responsible thing to do.

Then they confidently submitted their brief to federal court.

All case citations? Completely fabricated.

None of them existed.

Justice Arran Gerrard was… not amused. He referred the lawyer to the Legal Practice Board and hit them with over $8,000 in costs. His written judgment included a phrase I can't stop thinking about: "a good case undermined by rank incompetence."

In their affidavit, the lawyer admitted to developing "an overconfidence in relying on AI tools" and making "an incorrect assumption that content generated by AI tools would be inherently reliable."

Translation: they forgot that checking whether cases actually exist is kind of fundamental to being a lawyer.

As you might have guessed, this isn't really a story about AI failure.

It's a story about human failure to understand what AI actually is and how to use it.

The Problem Isn't the Tool—It's the Failed Collaboration Model

That lawyer didn't fail because they used AI. They failed because they used AI wrong. They treated it like a research assistant with a law degree instead of what it actually is: a pattern-matching system that sometimes makes things up.

They had no collaboration model. No framework for what AI should do versus what they should do.

And that's the gap I keep seeing everywhere—not just in courtrooms, but in boardrooms, HR departments, and strategy sessions.

So let's fix that.

Three Ways Humans and AI Actually Work Together

Different types of work need different collaboration patterns. One size fits none.

Model 1: Augmented Creativity

This is for open-ended, exploratory work where meaning and interpretation matter most—strategy, problem framing, content creation, sensemaking.

The human owns direction, meaning, and judgment. AI expands thinking, generates options, challenges assumptions.

In the failure mode, you'd be treating AI output as answers instead of stimuli. When leaders start copy-pasting AI strategy decks without critically engaging, augmented creativity becomes automated mediocrity.

Model 2: Hybrid Decision Systems

This is for decisions made under uncertainty where trade-offs matter—hiring, promotions, resource allocation, prioritization, risk assessment.

The human integrates context, makes trade-offs, owns the final decision. AI provides analysis, surfaces patterns, simulates scenarios.

The failure mode is silent automation creep. Leaders say "we still decide"—but somehow always follow the AI recommendation. The human becomes a rubber stamp.

Model 3: Oversight-Driven Automation

This is for repetitive, rule-based work where consistency and efficiency are primary—reporting, monitoring, repetitive workflows, rule-based execution.

The human defines boundaries, oversees performance, handles exceptions. AI executes consistently and scales efficiency.

Giving autonomy without oversight will lead to failure. And as our lawyer friend demonstrated, that can be catastrophic.

These models tell us how humans and AI work together. But they don't tell us who owns the decision.

For that, you need what the Stanford Human Agency Scale—which I covered in last week's issue. If you missed it, go back and read it. It matters.

Where to Go Deeper

I recently sat down with Scott Ritzheimer on The Secrets of the High-Demand Coach podcast to talk about building human-AI augmented teams. We got into the practical stuff—what works, what doesn't, and why most organizations are setting themselves up for the lawyer problem.