Shift

A friend recently shared two articles with me, Matt Shumer's warning about AI disruption and Connor Boyack's optimistic counter, and they perfectly capture the two dominant reactions to what's happening right now. The truth probably falls somewhere in the middle, though I lean optimistic.

The Shift is real

Since Anthropic launched Claude Sonnet 4.5 in September (and Opus 4.5 in November), something fundamental changed. The capabilities crossed a threshold that shifted how I, and many of my colleagues, spend our time. We often find ourselves directing LLMs to write code for us, rather than using it as a faster autocomplete.

What fascinates me most is the adoption pattern. Early on, it was the classic early adopters: junior engineers trying every new tool. Then, when the models dramatically improved, senior, staff, and principal engineers suddenly found themselves pushing code faster than they ever had in their careers. Now? Designers, marketers, members of our legal team are all using or interested in these tools.

As I write this, I have Claude working in the background on a spontaneous idea I had for work. I also have it rebuilding my personal website so I can have a markdown-powered blog, specifically to share this post. These aren't hypotheticals. This is my Friday evening, and I'm genuinely excited about it.

The challenges we can't ignore

The first article is right: there's nothing we can do but ride this wave and adapt. The potential upside is enormous, but several critical challenges I believe need solving before we get there:

Models are frozen in time. They can't truly learn during inference. There are workarounds that simulate learning, but they're not actually updating their understanding. This limits adaptability, but only for now.

Security, as always, needs thought. Prompt injection, data poisoning, and guideline violations remain unsolved. As these tools gain autonomy, the attack surface grows. My colleagues and I have written about practical ways to navigate these risks. The best thing you can currently do is control the data given to agents and/or limit the tools they access.

The economics don't add up yet. Not many people discuss this in the hype threads, but trillions of dollars have been invested in AI infrastructure. These models aren't running at cost. Data centers are expensive, energy consumption is staggering, and the compute requirements keep growing. At some point, the business model needs to work without infinite VC funding.

Governance and ethics are lagging. Who owns the output? Who's legally responsible when things go wrong? Who's accountable when they go right? Who captures the wealth generated? These aren't just philosophical questions, they're practical ones that will determine whether this technology benefits everyone or concentrates power and wealth even further.

The displacement question

Here's where I part ways with pure optimism: it's the breadth and speed of this change that's unprecedented.

Every previous technological revolution displaced specific categories of workers who could retrain for adjacent roles. Factory workers (eventually) became office workers, etc.

AI is different. It's a general-purpose substitute for cognitive work that improves across all domains simultaneously. When a significant portion of the workforce becomes economically unproductive faster than they can upskill, we'll need new answers.

This could genuinely challenge modern capitalism's core assumption: that most people can trade labor for income, which creates demand for more labor. Whats the solution? Universal Basic Income? New categories of human work we haven't imagined? A renaissance of art and creativity? I don't know the answer, and I'm not sure anyone does.

But here's what I do know: the cost of making things is plummeting. The barrier between having an idea and building it is disappearing. And I find it exhilarating.

So what now

More-so now than ever personal creativity and ingenuity matter. How can we harness these tools to build things 100x faster while injecting our own vision? (See my colleague's recent project, workflow.live, for a perfect example.)

The people who will thrive aren't the ones who resist this shift. They're the ones who engage directly, understand the limitations, and figure out how to use it to build things that matter.

I'm certain of one thing: we're living through a genuinely historic moment. The best thing anyone can do is push these tools beyond what you think they're capable of. More often than not, you'll discover they far exceed your expectations. And if something isn't possible today, there's a good chance it will be tomorrow.