The code was always the easy part

So, this is my first blog post … ever. I’d rather be building software than writing about it - but my recently-graduated-from-college daughters convinced me that some of what I’ve learned might be worth sharing. So here we are.

I’m a software engineer and the founder of a custom software development shop. I’ve spent decades studying, learning, and applying my craft to build really good software. My team of fellow engineers does the same.

When everything changed

We were early adopters of AI-assisted programming. At first, it was a somewhat annoying, over-eager coding assistant in VS Code that would pop up inane suggestions while the real expert (me) wrote elegant, performant code. Then it became a cool way to generate user interfaces from Figma designs - admittedly sloppy code, but the visual result was pretty good.

Then November 2025 happened. Anthropic released Claude Opus 4.5, which supercharged Claude Code … and everything changed.

Suddenly AI was capable of long-running, multi-step tasks and was outputting code that actually made sense. Given a proper spec and process to follow, it could create in hours what used to take days or weeks. It hit me all at once: humans are toast. Nobody is going to have a job in the near future.

The stages of AI grief

I’ll admit I went through what I’ve come to think of as the five stages of AI grief.

Shock. I’d watch Claude Code chew through a multi-step task and ship working code, and just stare at the screen. The craft I’d spent decades sharpening was suddenly available on demand.

Fear. Not for myself - for the young engineers on my team, and for my own kids. How does anyone become a senior engineer when AI writes the code juniors used to learn on? And how does anyone land that first job in the first place?

Depression. A stretch of “what was the point?” - not fun, and I suspect a lot of senior engineers are going to go through some version of it.

Acceptance. This is real. It’s not going back in the bottle, and pretending otherwise isn’t a strategy.

Hope. Eventually I stopped catastrophizing and started looking at what had actually changed. There’s real reason for hope here. Some real pain to work through, too.

The elephant in the room

Software development just became way easier. Given a proper specification and the right process - spec, build, verify/test, deploy - AI can build software at incredible speed. Writing code used to take 70% of the time and specification took 30%. That ratio is now reversed.

That’s uncomfortable for an industry that estimates work in engineer-months, hires on coding ability, and grows seniors by having them write a lot of code. None of us has a clean answer yet for how to scope new work, how to train juniors, or what “senior engineer” even means five years from now. I don’t either. But pretending the math hasn’t changed isn’t going to age well.

Why I’m hopeful

So why am I hopeful? Because the opportunities to create at unprecedented speed are amazing. We can help a client bring an idea from concept to production application in a fraction of the time. We can automate our own workflow in ways we never had time to before. We can build better software faster.

Someone may say, “but I can vibe-code an app myself.” Sure they can. I can also run my tax and legal documents through Claude or ChatGPT - but I’m not firing my accountant or lawyer anytime soon.

Building real software is the same - most of the work is invisible. Security, data privacy, performance under load, reliability, the long tail of edge cases that show up the day a real user touches it. None of that shows up in a demo; all of it decides whether the app holds up when it leaves your laptop.

I trust human subject-matter experts to guide me in areas where I’m not an expert. Will they be using AI to help perform their services for me? I assume so. But they actually know what they’re doing in their field, and AI is a force multiplier for them, not a replacement.

What AI can’t do

AI in software development specifically looks scary because AI is really good at writing code. But it’s not good at sitting down with a client team and working out the details of a project. Or choosing the right approach for the client’s situation - now and for the future. Or recognizing when the “obvious” solution is the wrong one.

That’s the work. And that’s the work that matters most.

The code was always the easy part - we just didn’t know it until now.

Where I’ve landed

AI doesn’t replace experienced software teams. It makes the good ones dramatically more valuable. We can take a client from concept to working product in weeks instead of months. We can prototype three approaches in the time it used to take to build one. We can spend more time on the conversations that actually move the project forward, and less time on the plumbing.

If you’re an executive sitting on an idea you’ve been meaning to build, the math has changed. What used to be a six-month commitment can now be a six-week experiment. That’s the opportunity I’m hopeful about - not the replacement of humans, but the unlocking of ideas that were too expensive to try before.

Getting there takes humans staying in the driver’s seat - bringing judgment, taste, and care to the problems worth solving - while AI handles the parts it’s genuinely good at. Keep humans in the loop, use AI as the force multiplier it actually is, and the next decade gets bigger, not smaller.

That’s the work I want my team to keep doing. And the future I want my kids to step into.