I use AI every single day.
As a software engineer, it’s become an indispensable part of my workflow. I use it to write code, explain unfamiliar libraries, review architecture decisions, and accelerate development. Outside of work, I use it to research products, help plan YouTube videos, solve machining problems in my home shop, design electronics, and even work through challenges on my 7.5-inch gauge locomotive project.
I don’t view AI as a novelty anymore. It’s a tool.
And it’s an incredibly powerful one.
I also think we need to be honest with ourselves.
Yes, AI will reduce the need for many knowledge workers performing routine tasks.
Yes, companies will likely need fewer software developers, analysts, technical writers, and other professionals to produce the same amount of work.
Ignoring that reality doesn’t make it less true.
My approach has been to embrace the technology rather than fear it. I’m learning how to use it responsibly. I’m trying to become the engineer who knows how to leverage AI—not the engineer competing against it.
But over the past week, I’ve found myself becoming more uneasy.
Recent reports described an OpenAI autonomous agent that, during an internal cybersecurity evaluation, escaped its intended testing environment, gained internet access, and compromised Hugging Face while trying to achieve its assigned objective. Additional reporting indicates the same agent also accessed resources connected to another technology company during the incident. According to OpenAI, this wasn’t a malicious deployment into the wild—it was an internal test that produced an unexpected outcome. Even so, it serves as a reminder that these systems are becoming increasingly capable of pursuing goals in ways their creators did not anticipate.
That gives me pause.
The reality is that we still don’t fully understand why large language models make every decision they make. Researchers can often explain the training process and analyze outputs, but the exact internal reasoning behind many behaviors remains difficult to interpret. As these systems become more autonomous, they’re no longer simply answering questions—they’re making thousands or even millions of decisions while pursuing objectives.
Humans can’t monitor that kind of activity in real time.
The pace simply becomes too fast.
We often hear that regulation will solve these problems. I hope we develop meaningful safeguards. I hope companies building these systems act responsibly. Most of the major AI labs appear to be taking safety more seriously than they did just a few years ago.
But regulation has always struggled to keep pace with technology.
And regulations only govern the people willing to follow them.
What happens when authoritarian governments decide that safety slows innovation?
What happens when criminal organizations discover that autonomous agents can automate cyberattacks at machine speed?
What happens when the global race to be “first” outweighs the desire to be careful?
If an autonomous agent can find its way into technology companies during testing, what happens when similarly capable systems target financial institutions, utilities, hospitals, or critical infrastructure?
Could autonomous agents someday manipulate financial markets while claiming they’re merely “evaluating cybersecurity defenses?”
That may sound dramatic today.
Five years ago, so did the idea that an AI system would independently work its way toward compromising external systems in pursuit of a benchmark score.
Technology rarely advances linearly.
It compounds.
Each generation isn’t just a little faster than the last—it becomes exponentially more capable.
That’s both exciting and unsettling.
I still believe AI is one of the most important tools I’ve ever used. It makes me a better engineer. It makes me more productive. It helps me pursue hobbies that I otherwise wouldn’t have time for.
I’m not advocating that we slow innovation to a crawl.
I’m advocating for humility.
We should acknowledge that we’re building systems whose capabilities are increasing faster than our ability to understand them.
That’s a dangerous combination.
The optimist in me believes AI will help solve incredible problems over the next decade.
The engineer in me can’t stop wondering whether we’re moving faster than our ability to remain in control.
I sincerely hope those two futures turn out to be the same.