Engineers and companies need to realize that we are now holding agents back

In November/December of last year, the AI models broke a threshold. There were several things that converged that made the end of 2025 the pivot point.

It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December --Andrej Karpathy, feb 26 2026

Karpathy called it a phase change - not an incremental improvement. He argued that coding agents crossed a qualitative threshold since December. From brittle demos to sustained, long-horizon task completion with coherence and tenacity. And even though things were absolutely moving fast up until then, crossing this threshold meant things would move exponentially faster from that moment on. And they did. So much has happened just these last 3-4 months, and things are NOT slowing down.

Crucially, this means that agents are now held back by us. It's what Karpathy calls a "skill issue". We're now in a situation where the agents are very often wrong, but this time they are wrong because of us. The technology is moving so fast that we are being left behind.

With the right harness and the right instructions, agents are now capable of a lot more, but most of us simply don't have the skills yet. We're all scrambling to keep up, trying to follow the pace of development, but there is just so much going on. So many people who are doing incredible things everywhere, and it is information overload. Companies today need skilled people to help steer them through uncharted territory.

So, if you are an engineer, you really need to step up your game in learning all these new shiny skills that make up the ultimate skill right now: controlling agents.

And if you're a business leader, you need to realize that this is the moment where you'll have to invest in your employees more than ever before. We all need to learn new skills so we can harness this new technology. Don't even consider how AI might save you money by solving tasks that you are paying employees to solve today. Is your goal to continue your previous trajectory with fewer people and lower costs?

Consider instead what you might suddenly be able to achieve when your employees have access to so much power!