The Opus 4.5 moment isn't just about a better model

Something changed in the water in December. Yes, it's a big deal that some of best engineers have flipped the way the code to agent first. 10x engineers might have gone to 100x. But, for us 0x engineers, the change is arguably even more seismic.

So what changed? Yes, Opus and Codex have gotten a whole lot better. But it's not just model quality that changed, but tool reliability. The models of two years ago were already pretty good, but they were idiot savants. We've been talking about "vibe coding" for a year now, but even 3 months ago it required a whole lot more hand holding and troubleshooting. The problem was the memento challenge - highly capable models, but with short attention spans, limited memory, and easily distracted.

What slowly evolved over the last few years was not just capable agents but agentic loops. Take a task through a series of specialized sub-agents with focused context, maybe even just make the agent keep trying until it gets it right, and suddenly you get a much more reliable result. The memento meme became a series of specialized polaroids that meant the agents didn't just wander off but consistently accomplished what you set them to.

A lot of these solutions started as 3rd party libraries or skills, but once proven they got incorporated into the platforms as native capabilities. It's not just that Opus got a whole lot better, but Opus in Claude Code did. Features like plan mode, memory, security review, swarms all turbocharge an already capable model into a truly transformative platform. "The hottest new programming language is English" was tongue in cheek 3 years ago but is just true now.

So... what comes next? Many of these are not groundbreaking conclusions, but are still worth reiterating:

1. Writing code is about to become commoditized. Not software engineering—the judgment, architecture, requirements definition, edge case anticipation. But translating requirements into syntax? That's mostly the model's job now.

2. MVPs will now come much fully featured. When the cost of building drops this much, the math on smaller iterative discovery doesn't make the same sense.

3. That's going to evaporate a lot of moats. Distribution advantages, network effects, proprietary data still matter. But there's a whole lot of SaaS that differentiates most only feature set. That advantage? Gone.

4. A whole lot of business models are about to get incinerated. When you've built up a team around being able to charge one price point and suddenly a competitor can offer the same features for 1/10th the price, it's going to be hard to adjust and many wont.

5. Software is going to get a lot more personal. Customization that might have required 6 figure contracts in the past might start making sense for a $10k customer. Hobby projects will be a lot more common and anyone who is even remotely technical will have at least a custom tool or two. Some products will incorporate that personalization natively giving more flexibility to adjust your UI as desired. Others will pivot to providing pipes that get personalized to the user by their own agents.

6. In hindsight, it makes sense that coding was first. There's a lot of programming thats just researching, pattern matching, and applying patterns in the right syntax. It was a perfect match for RLVR + agentic loops. There's a whole lot of other fields that could apply to and the race is on be first in building all the different Claude Code for [X]s (or perhaps more aptly to beat Anthropic there). What's new is much clearer now how far it's possible to get in an extremely short period of time.

7. This year will be rough on the job front. Last year we didn't see too much in the AI-catalyzed layoffs, though we did see some signs of slower hiring. That won't be true in 2026.

8.The societal impact of that will be unimaginably large. The impact offshoring had on manufacturing fundamentally reshaped global politics, and it happened on the time scale of decades not months and years. This shift will be faster and impact a lot more of what we now call the "good" jobs.

It's hard not to feel both the incredible potential and terrifying consequences this moment holds. I'm neither a blind optimist or cataclysmic pessimist. But I am quite shaken for better and for worse.

AI doesn't need to reach AGI to change everything. We're only feeling the first foreshocks now and it already feels like the ground is moving beneath us. The actual earthquake is going to be several orders of magnitude larger and it's just around the corner. Buckle up while you still can.