I’m having more and more conversations lately with folks - Engineers, Product Managers, Designers - who are all feeling a mix of fear, anxiety and malaise about the rapid advancement of AI coding tools. I’ve been struggling to get to the heart of why this feels like the wrong reaction to this moment, and to articulate why I’m still optimistic about the future.

An article that has struck a chord with some folks recently is Steve Yegge’s Welcome to Gas Town. In it, Yegge paints a picture of a development approach where Engineers act as managers of a swarm of AI-coding agents; directing work, reviewing outputs, and keeping agents unblocked, all while the actual code, reviews, and implementation details are abstracted away (for the most part). In response to this type of thought leadership, Software Engineers begin to tense up and ask “What does this mean for my job over the next few years, if I’m no longer writing or reviewing code?”.

Code is rarely the bottleneck

Kellan Elliott-McCrea recently published a great post called Code has always been the easy part. I loved this bit:

“…successful teams have always known that the value is the system, the value is the human-technology hybrid that allows a product to be delivered, meet customer needs, evolve to provide more value over time, meet the spoken and unspoken needs of the problem domain…”

Here’s my framing of what Kellan is saying: Code is rarely the bottleneck. Let me be even more specific: Code is rarely the bottleneck to producing true, lasting value/revenue. Features can often be a bottleneck, but we’ve been able to hustle toward crazy deadlines way before these AI tools existed. For a long time, it has been easy to ship a ton of code quickly, when that is what you want to do. In my time working with and managing agile teams in larger orgs, we’d of course always like to move more quickly. But the more common and vexing constraint is knowing what to work on next, ensuring we have clear requirements, and genuinely understanding the problem at hand.

The most meaningful bottlenecks and most difficult work have regularly been things outside of writing and reviewing code. If you’re building software, I imagine you’ll agree: we spend far more time figuring out what to build and how to build it than we do actually building the thing. The rapidly advancing set of AI tools add yet another abstraction layer for engineers, and make the build step much faster. I think that’s awesome, because it allows us to focus on defining what to build and how it should be architected. We can afford to get better at this.

Software is rarely a moat

Another acute version of the collective anxiety has taken this form: the idea that someone else will clone your company’s software in just a few weeks and come to take all your business. In my experience, software is rarely the moat. If your software is the only special thing about your company, then your company might not be all that special. When I look back at my experiences in Enterprise SaaS, we were always building our version of a CRUD app that was ultimately trivial to imitate. There was always a parade of imitators, and none of them ever made a meaningful dent. Customers buy our reputation, our brand voice, and our POV on the space alongside the software. Software is roughly as easy to clone now as it was before agentic AI; the business that sells, supports, and stands behind it is significantly more difficult to clone.

Speed of innovation still matters - especially between two closely matched competitors — but only if you’re shipping the right thing. We have to assume our competitors are using these tools to stay efficient. If your team is spending 25% of its capacity on work that could be automated, your competition is going to innovate faster. The playing field is leveling and if you’re not experimenting with these tools at all, your anxiety is not misplaced.

Why I’m still excited

It might not be coming through, but I remain genuinely optimistic about what these tools can do for us. It’s the negative reactions that feel off. The question “when will AI be able to do 100% of my job?” sells us all short — and I think it’s the wrong question entirely. Maybe this is a better line of questioning: if you used to spend 25% or more of your time literally typing out code, what are you going to do with that time now? I’ll hazard a guess:

• You’re going to spend quite a bit of it prompting agents, and eventually orchestrating work using a bunch of different tools in order to make light of (or completely automate) most coding tasks. For a lot of folks this may take just as much time as it does to write code today. Whether the overall output is greater isn’t really the question, since “lines of code” has never been a good measure of value or productivity.

• You’ll spend a non-zero amount of time hand writing code in situations where it’s still far easier to do a thing than to explain to an agent how to do a thing.

• Ideally, you’ll use the time to do better research about what to build, think deeply about how it should be architected, and deliver solutions you’re proud of.

We’re all a little worried about job security, and that’s probably fair. Over time, the rise of these tools may mean that companies can operate with far fewer people. I’d argue that it’s just as logical that many more people can start software companies too. If you’re still asking “so what?”, I’ll leave you with this: if you can focus your energy on using today’s tools to deliver real value rather than give into anxiety, you just might have a reason to be optimistic.