Should Software Engineers Be Anxious About AI?

In September, AI agents (AI models that can use tools and take actions on their own) produced machine-checked proofs on part of the Navier-Stokes¹ problem, a question about the equations that describe how fluids flow, open for roughly 90 years. Similarly, Anthropic reported² that roughly 950 agents spent 21 hours searching a database of DNA sequences and found a previously uncharacterized enzyme system, with the involvement of its scientists limited to the initial prompt and the lab work. There are also many reports of teams shipping complex, production-grade code written mostly by agents, such as GitHub's rewrite of its Copilot runtime in Rust³, and at Rails World, DHH, the creator of Ruby on Rails, said⁴ that writing code by hand "is no longer an economically viable skill for most programmers at most companies." Unfortunately, this has had the effect of driving up AI anxiety. By that I mean the career kind: the worry that AI will soon do the work you spent years learning to do.

So, should you be anxious about AI? I have lived through a few revolutions (e.g., the mobile first⁵ shift, the rise of wearables, the software 2.0⁶/deep learning phase), and my assessment is that anxiety, while reasonable, can be managed. Specifically, what has worked for me is accepting that the nature of work has changed, and meeting the moment with agency, creativity and curiosity.

In the rest of this post, I reflect on what is driving the anxiety, and what you can do to adapt.

I Understand the Anxiety

As engineers, we have all taken pride in a deep understanding of how software programs work. We have invested years in understanding distributed systems infrastructure, web application design and interface design. Today, frontier AI models have hoovered up all of this knowledge at an unimaginable scale, and paired with a good harness (the tools and instructions around the model), agents arrive at solutions that are often not far away from what a good professional would end up with. As this all gets better, it forces the question: what happens when AI can accomplish all that I have spent years learning, at a speed and magnitude I can never match? This is what drives anxiety. Software engineer Obie Fernandez⁷ has a name for the fear: being a meat proxy, someone whose job is to "fit into a process where other people decide what to do" and then "transform those directions into software."

Related article

You have 'AI Fatigue' - That's Why You Feel Awful

A reflection on the mental toll of keeping pace with AI's rapid advancement and how it affects researchers, engineers and organizations.

newsletter.victordibia.com

What Should You Do?

People often ask me which skill to develop so they stay relevant, or how to respond to the anxiety above.

I think the first step is accepting that the work has changed. Then practice three habits: agency, creativity and curiosity. Across the shifts I mentioned above, these are what helped me the most.

Accept That the Work Has Changed

Personally, I think this will be the hardest part. For the last 20 years, what a software engineering contribution means has been standard: putting skill into writing reliable software that does not break. Today, agents can write most of that code, including the parts that make it reliable and let it scale. It's a big change. There will still be work to do, but it is different work. Stephen Toub, a Distinguished Engineer at Microsoft who drove GitHub's Rust rewrite of the Copilot runtime³, describes his new role: "instead of being responsible for writing syntax, I was responsible for framing the problems, defining boundaries, choosing strategies, adjudicating exceptions, and overall ensuring everything was moving in a good direction."

Agency

I think the key point here is a bias towards action. Psychologist Albert Bandura defines it this way: "To be an agent is to influence intentionally one's functioning and life circumstances"⁸. In practice, agency is a loop: observe, act, learn from the results, and act again. Failures along the way are fine; what matters is that each attempt is intentionally tied to some form of learning. Most decisions are what Jeff Bezos calls two-way doors: reversible, so you can reopen the door and go back through⁹. Agency also means owning the outcome. When you ship software that people use, things will break, and we still need accountable individuals. Katie Kirsch¹⁰ of a16z has a simple version of this: tell people what you're going to do (mostly your team or organization), do it, and tell people you did it.

Creativity

I tend to think of creativity as coming up with new good ideas. Creativity can also mean connecting new dots across existing ideas. Cognitive scientist Margaret Boden calls its most common form combinational creativity, unfamiliar combinations of familiar ideas¹¹. AI models are now one of the ingredients you can combine to produce results that were not possible before: two days after Claude Opus 5.5¹² came out, Victor Mustar, head of product at Hugging Face, combined it with LEGO design¹³ to produce a buildable model, complete with an animated visualization and a 141-page instruction booklet, which he says the model made. I think this shows how the starting point (the floor) of what counts as creative or interesting has risen for all of us.

Curiosity

Curiosity is occasionally coming up for air to sample the state of the art: what the tools can do now, and what others are doing with them. If there is a new model, new harness, new compute substrate, new architecture: read up on it, try it, and test out your agency. Apply agents to your daily life too. It keeps your mental model of what they can do current, and the habits it builds carry back into your job. Curiosity also builds skill: getting fast and efficient with agents, so that using an agent doesn't mean one step forward and three steps backwards.

You Need All Three

In addition to acceptance, I think that to truly thrive in the new world, you do need to practice all three aspects in good measure.

• Without agency: agency is where it all begins. A belief that you can act and that your actions can move you closer to desirable goals. Frankly, this was also a required ingredient for success before the AI world, but more so today.
• Without creativity: you're doing what everybody is doing. You're not asking what happens if you combine the agent with some new thing.
• Without curiosity: you are stuck in a snapshot in time. You're acting, connecting dots across domains and coming up with new ideas, but the tool you're working with has changed, and so has the set of applications people are applying it to. A year ago, it was preposterous to talk about an AI model creating an application launch video; today, I'm confident anybody creating one should begin by asking a frontier model to do it. Being curious means continuously evaluating the state of tools and practice across the industry.

Together, these three habits are what keep you from being a meat proxy.

How I Feel About It

I miss the times when the pace was slower, when over the course of months I could noodle on a problem, or take weeks to write a blog post to think through an idea. One consequence of how fast agents work today is that I am constantly becoming the bottleneck, leading to new tensions (should the code be read, should we focus only on verification, how much should we trust our software factories). I have accepted this new way of working, but I am also focusing on ways to ensure that while velocity improves, quality still holds.

Further Reading

• Tim Dettmers, dlab Open Source Week: Frontier AI on Your Own Hardware (September 2026). Written to students and PhD candidates who are anxious about AI: "The difficulty did not disappear. It moved."
• Obie Fernandez, You're Already a Meat Proxy (September 2026). The essay the term "meat proxy" comes from.
• Stephen Toub, Migrating the GitHub Copilot runtime to Rust, using Copilot (September 2026). What one engineer's job looked like on an 800,000-line rewrite written mostly by agents.
• George Mack, High Agency. A short essay on agency as something you can practice.
• From me: How will AI Impact Software Engineering? (2024) and How will AI Impact Academic Research (and Publishing) by 2027? (2025).

References