推动 AI 智能体走向实用

At the end of last year, AI agents really came alive for me. Partly because the models got better, but more so because we gave them the tools to take their capacity beyond pure reasoning. Now coding agents are controlling the terminal, running tests to validate their work, searching the web for documentation, and using web services with skills we taught them in plain English. Reality is fast catching the hype!

This is all very evident if you've tried to employ any of the new models — especially Claude Opus 4.5, Codex 5, Gemini 3, and even the Chinese open-weight models like MiniMax M2.1 and GLM-4.7 — in one of the modern terminal harnesses that give them access to all these autonomous powers. The code being produced by this new breed of AI is leagues ahead of where their predecessors were at the beginning of 2025.

I've thoroughly enjoyed putting them all to work in OpenCode, which is a terminal interface for coding agents that allows you to seamlessly switch between all of the models, capture your sessions for sharing, and simply looks astounding when theme-matched with the rest of Omarchy (where we're making it a default in the next version!).

See, I never really cared much for the in-editor experience of having AI autocomplete your code as you were writing it. That was the original format pioneered by GitHub's Copilot and Cursor, but it left me cold. When I code, I want to finish my own thoughts and sentences. That was the sentiment I expressed on the Lex Fridman podcast last summer.

But with these autonomous agents, the experience is very different. It's more like working on a team and less like working with an overly-zealous pair programmer who can't stop stealing the keyboard to complete the code you were in the middle of writing. With a team of agents, they're doing their work autonomously, and I just review the final outcome, offer guidance when asked, and marvel at how this is possible at all.

Yes, I'm ready to give the current crop of AI agents a promotion. They're no longer just here to help me learn, answer my questions, or check my work. They're fully capable of producing production-grade contributions to real-life code bases. 

Yet pure vibe coding remains an aspirational dream for professional work for me, for now. Supervised collaboration, though, is here today. I've worked alongside agents to fix small bugs, finish substantial features, and get several drafts on major new initiatives. The paradigm shift finally feels real.

Now, it all depends on what you're working on, and what your expectations are. The hype train keeps accelerating, and if you bought the pitch that we're five minutes away from putting all professional programmers out of a job, you'll be disappointed.

I'm nowhere close to the claims of having agents write 90%+ of the code, as I see some boast about online. I don't know what code they're writing to hit those rates, but that's way off what I'm able to achieve, if I hold the line on quality and cohesion.

But I'll forgive folks for getting excited! Because you don't have to connect many future dots on the current trend line to get dizzy by the prospects. The leaps of improvement that AI agents took in 2025 is simply incredible. This is the most exciting thing we've made computers do since we connected them to the internet back in the '90s. So what might things look like in 2026 or 2027? I get the exuberance.

I also get that some programmers are eager to tune it all out. The hype drones on relentlessly, the most fantastical claims are still far off from being substantiated, and there's real uncertainty about where all this will leave the profession in the future. But that's still not reason enough to miss out on this incredible moment in human and computing history! 

You gotta get in there. See where we're at now for yourself. Download OpenCode, throw some real work at Opus or the others, and relish the privilege of being alive during the days we taught the machines how to think.