Over the last few months Large Language Models, and the (misnomer?) colloquialism of "Artificial Intelligence" have occupied waking thoughts with, fears, predictions, and awe. This is the first time in my decade of working in software, and generally speaking my life, in which I have felt a paradigm shift so significant and so life changing that it feels as though Pandora's Box has been opened and my career as it was will never be the same.

Delivery in the Age of the Agent

The act of writing code is fun

I've always enjoyed the physical act of penning code to my editor. I use NeoVim. I rely on a custom keymap for my Moonlander. I'm obsessed with shortcuts and optimizing my workflow for fast editing.

Now, in the "age of the agent" I am spending most of my day prompting agents instead of navigating brackets, changing words or coming up with the ultimate codemod or vim macro to do my bidding.

And, to be honest, that is depressing.

Taking a step back however, it is undeniable that I'm delivering more software, of higher quality, and faster.

So why does delivering value faster with a prompt feel worse than delivering value faster with learning a new vim keystroke?

That is much harder to define, but for me, it simply has to do with the user experience of typing in a chat window versus using vim motions to navigate a project. (I also believe this is why I've gravitated towards Claude Code)

It's an undeniable fact, however, that as long as this genie is out of the bottle I will need to use AI to stay competitive with my peers. But there is hope.

Delivering value with software is fun

Delivering new features, User Interfaces, APIs and business value is still, in my opinion, fun and satisfying. In the "age of the agent" that feedback loop is becoming tighter and tighter.

When I have an idea for something new, suddenly the coding part is commoditized. I can go from idea to delivery much faster, of (usually) higher quality and that is fun.

On Quality

Quality is a tricky thing to talk about when it comes to AI. The LLM Luddites will tell you no computer generated code will be anything but a poorly performant, derivative mess. However the AI technophiles will espouse its machine-like perfection, attention to detail, and limitless work ethic.

The reality? As with all things I believe it to be somewhere in the middle. More importantly, AI is a tool, and a good one at that. Like all tools it can be used well, and it can also be used poorly.

I've found that using this tool results in higher quality diffs that I'm submitting to my peers. Why? That's easy. I've never considered myself a technically strong software engineer. I've never fully understood TypeScript decorators, or the intricacies of useEffect. An LLM is very good at handling the rote tasks that are easy to miss. LLMs rarely make syntax errors, they can easily ensure that desired functionality matches observed functionality. All things that humans, especially in the time crunch that the corporate world can be, can easily miss.

Attention to detail has never been my strong suit, and it's something I have to work on constantly. LLMs have allowed me to pair with "someone" with an incredible attention to detail, and I get to focus on ideation, certifying functionality and delivery.

The Death Throes of "Coding"

The days of penning code to an editor "by hand" (i.e. "coding") are almost assuredly numbered. Syntax dominated by curly brackets and semicolons is yielding to new regime of natural language. We will undoubtedly miss the old paradigm, and that's okay. Plenty of coachmen missed their carriages.

This new world is one of commoditization. It's easy to feel overwhelmed by the rate of change. It requires constant research and learning. Thankfully, however, being a technologist has always required and rewarded these things.

We'll be done coding, long before we're done delivering software that brings value to people's lives.

It may not always be the case, but for as long as delivering value with software is fun, we'll all be just fine in this new world. Even if we're not coding.