I don’t know about you, but lately I’ve been sensing a lot of anxiety about how AI is going to replace us. Software engineers are posting worried takes all over LinkedIn and in private forums, while product-focused people understandably enjoy the feeling of empowerment a bit of vibe-coding can give them. I’ve even had a few people book coffee chats or request 1-on-1s to vent, discuss their approaches, and compare notes.
Personally, I’ve been using AI for code and research extensively for about two years now—not long enough to be called an early adopter, but long enough to feel the change under my feet. I’ve built AI integrations and simple yet sharp tools for my work with it, and I have a mostly positive impression of the impact on my work, broader societal aspects notwithstanding1.
I’m not worried AI will replace me. Not because I think I’m a special snowflake with irreplaceable skills, but because I don’t think our value was ever in pushing out the biggest quantity of code the fastest. If someone’s value is primarily shipping code for very well-defined issues, then yes, that part is now cheaper. But that’s not our craft, even if it’s the most visible output, and a big part of our identity.
Today I want to talk a bit about how I think about questions of AI and replaceability, and what that means to my craft—productivity, the way my role is shifting, and how I’m coming to terms with a big pillar of my identity changing.
Purely producing good code has definitely taken a backseat lately. I still practice my craft, and I still enjoy doing it. But for a lot of everyday tasks, I take the productivity gains of prompting a model well, giving it access to my tools, and then refining and iterating as needed. It’s fun, though definitely in a different way than plopping into Vim (still my editor of choice, 15 years and counting) and banging out lines and lines of code to solve the problem.
But we do solve the problem regardless, and in a professional setting, that’s honestly just as fun for me.
The questions that I didn’t have to ask myself, and that I hear being tossed around a lot, are questions of identity, belonging, and purpose. We belong to a tribe, and our tribes were until very recently shaped by our technological choices. Now that matters less and less, and what matters more and more is the product aspect, and maybe that’s just not what some of us signed up for, what gets them up in the morning, what grips them. That’s fine. I feel lost sometimes, too. But we’re not immutable entities, and we can find meaning and purpose in lots of different things.
If a lot of your identity is tied to the visible part of the work (shipping code), then “code becoming abundant” can feel like having to face replaceability. I don’t think that’s an inevitable conclusion, but I understand why it feels like that emotionally.
So now I also find purpose in building things that otherwise would have felt like too much. I can expand my project backlog even more. I shepherd hundreds of repositories all over the internet, some of them useful, some of them silly, and I care about most of them. It started to weigh me down a bit when it came to exploring new things, and I think AI helped me gain back a bit of momentum. It feels nice.
By understanding AI as a tool first and foremost, it loses a lot of its teeth. Because it is my tool, and I still wield it. It’s a skill that can be learned and that is useful, and I know how to and like the process of learning new things.
But what do I do with it?
I use AI as a companion that helps me build things quickly. It bootstraps CI, packages software, lays out documentation, and writes code. I also focus on writing code alongside it, but even more on understanding what it produces, and keeping the architecture reined in.
The components of a system and how they interact still matter. Finding the right abstractions and the design language of your environment still matters. As of now, I wouldn’t let AI create that part, and I take pride in still producing results that feel like me in a way, not just machine output.
And of course, on an even higher level, I still need to think about posture, about security, about responsibility, accountability, and the positioning in the wider ecosystem. These are decisions that I need to take, because I will be accountable. Abundance of solutions doesn’t remove decision-making, it makes it harder, and maybe even more valuable.
A couple of weeks ago, on a whim, I started an experiment under a moniker that I had created a long time ago. I’m going all in on AI-first development within the constraints of cyberwitchery lab, a cheeky little brand that I built for myself (for now without a legal entity behind it).
The premise is simple: I want to find out what still matters to me when raw velocity is less of a concern.
I’m starting with a few projects that I had in the back of my head anyway, and that I needed an excuse to work on. A timetracking app. CLIs and Rust API clients for Netbox and Nautobot, and maybe some other DCIM/IPAM systems. A tool to keep my prompts and agent definitions organized, composable, and lintable.
All of these I could have built by hand, but have now deliberately chosen not to, simply to learn how an AI-first workflow that is clean, safe, and produces real output with real value looks like for me. It’s a proper attempt at extending my skillset in a way that remains aligned with my values.
If you’re curious about the applied side, what “AI-first” means for my day-to-day, what tools I use as guard rails, and what I’ve learned so far—I’ll write that up over on the cyberwitchery blog soon.
Right now I’m working with Claude, Codex, Gemini, Claude+Ollama. Grok is on my backlog. I’m trying everything and seeing what sticks for me. I’m expecting this list to change as all of my other tools have over the years.
And even if my little lab ends up failing, I can say that it produced valuable output already. I really needed all of the tools it produced, and I think I’m not alone in that. I call it a lab deliberately, because it is allowed to fail. I don’t want to make a company, I want a space that I can build and validate or falsify in, and demonstrate what I actually want to bet on right now: AI is not replacing us, but it does change what engineering looks like.
I want to find out where we’re going2!
I don’t want to diminish your experience if you feel doubt or anxiety around the topic of AI taking our jobs—especially if you were affected by the weirdnesses of the current market. All I hope is that we can reframe the problem, and try to treat AI as a tool first, because that is what I see it as3. A tool that demands respect, time, and attention if it is to be learned, surely, but a tool nonetheless.
And I’ve never been one to pass up a good tool.
Thanks to Moritz Neeb and Kai Williams for reviewing drafts of this post!
1. I’m not blind to the broader societal aspects of AI, but I think there is room to discuss the tool on its own merits, especially as it pertains to what we do, without having to face the more unsavory aspects of big tech and corporate behavior in that particular conversation.
2. If you’re experimenting with this too, I’d love to compare notes!
3. Bold claims of people with skin in the game to make the “line go up” notwithstanding, I cannot survey the current market and see it as anything more than that. But as with bitcoin as a new currency, there are strong market forces at play here that want us to lose sight of this. I try to look at what is there, and not what is presented to us. You are of course entitled to an entirely different opinion, and I will let my mind be changed quite easily as new facts come in.