Last week, I had the privilege of sitting on an AI panel where the first question asked was, “What does AI mean to you?” My answer? “To me, AI is a kickass intern.” It’s eager, fast, and surprisingly capable at times—but it still needs oversight, coaching, and occasionally, someone to clean up its mistakes. Like any tool, it’s only as good as the guidance it’s given, and the wielder must know its limits.

This idea isn’t new—research and experts have long pointed out that while generative AI can make us more efficient, its value diminishes without human involvement. Both Aaron and Rachel have included this theme in their blogs, and respected institutions like Stanford and Forbes have published on this topic. Despite being well-understood, generative AI still makes obvious, stupid mistakes. If it were easy to fix, I’m sure OpenAI, Google, etc. would have by now, so folks - buckle up, it’s going to remain rocky for a while!  

To move from the abstract to something everyone can relate to, let me share a few examples from my personal life in which generative AI has made inexcusable mistakes, which highlight this truth.

When AI Fumbles the Basics

Here are three moments from last week alone that showcase how AI can stumble—even on the easy stuff.

Example 1: When AI knows the answer, but gets it wrong anyway
I asked an AI tool to calculate the nutrition for Benefiber. It confidently told me it was 5 calories per serving. When I corrected it, the AI backtracked and gave me the right number: 15 calories per serving. How did it miss something it knew?

‍

Example 2: When AI assumes you don't need the truth
Curious about timestamps, I asked the AI for the time of my last question. It told me 4:36 PM—problem was, it was only 3:06 PM. When pressed, the AI admitted it had given a hypothetical example without clarifying that upfront.

‍

Example 3: When AI forgets its own rules
While working on a muscle retention calculation, the AI used the wrong formula—even though it had applied the correct formula earlier in the same chat thread. Its error led to an impossible result, and only after I pointed this out did it acknowledge the mistake.

Tools Are Only as Good as the User

These examples are rather frivolous, low-stake scenarios but they underscore a fundamental truth: generative AI is just a tool, and further, one that’s hit or miss. It might be friendly, resourceful, and willing to jump in anywhere—but it’s also prone to misunderstandings, overconfidence, and the occasional rookie mistake. At the end of the day, it cannot replace the responsibility of its user (that’s you!) to review and validate its outputs.

Think of it like an old-school watch. If you’re trying to get to an appointment on time, the watch might be your tool of choice but even when it’s faulty you’re still the one accountable for arriving promptly. Whether you rely on a sundial, news radio, a digital clock, or your phone, you must choose the most accurate and reliable tool—and still double-check it when it matters.

The Stakes Are Higher in Engineering

This analogy extends to work, particularly in engineering teams adopting AI Code Writing Co-Pilots. These tools can be game-changing, but their outputs still require human oversight. The stakes are too high to blindly trust a machine’s confidence.

How Flux Fits In

This is where Flux comes in. Flux doesn’t just point out errors; it provides engineering leaders with the insights and clarity they need to take confident action. By centralizing information from multiple sources, Flux ensures you’re not just seeing the issues—you’re understanding their context, impact, and priority.

It’s like having an experienced guide by your side, helping you cut through the noise to focus on what matters most. Whether it’s identifying risks, prioritizing tasks, or providing actionable recommendations, Flux empowers leaders to rely on AI tools while staying firmly in control of the outcomes.

The Bottom Line

AI may be a kickass intern, but it’s your expertise—and the right tools—that turn its potential into results.