Table of Contents
In Search of Purpose
LLMs are all the rage nowadays, and as with any new technology, people are still figuring out how to integrate them into their lives.
It’s a deeply subjective experience, because different people want different things and have different standards. I’m focused on professional programming use cases, so I won’t expand on amateur “vibe coding” journeys or other funny ways people amuse themselves with LLMs.
First Gen
The first LLM products were chat‑based and weren’t embedded into codebases or IDEs, which introduced a lot of friction for programming tasks. They weren’t completely useless, though, and I found a solid use case for them: removing abstraction layers.
I have a few non‑programming use cases too, like cooking, checking grammar, and circumventing paywalls. It’s pretty cool to paste a headline and have the model fetch all the relevant data and answer follow‑up questions.
Needless to say, most productivity claims about LLMs are overstated. It’s nowhere close to 10x, and is often much lower than 1x. But that’s fine if you consider them entertainment or quality‑of‑life tools rather than pure productivity tools.
LLMs, when used carefully, can enhance code quality, which is arguably much more important than raw productivity, but quality is a hard product to sell.
Second Gen
OpenCode is an attempt to boost LLM productivity for programming tasks. You just need to navigate to a project folder and type opencode in your terminal. It has free rein in that folder, so it’s not unreasonable to expect productivity gains from eliminating copy‑pasting, which was the most annoying part of the old chat‑based workflow.
Even with OpenCode and the most advanced LLMs, I still don’t see any meaningful productivity gains. But that’s totally okay, because it helps me with dumb, time‑consuming, error‑prone refactorings and other easy, measurable, but labor‑intensive tasks.
Conclusion
OpenCode is a great tool for adding simple repetitive features or doing controlled refactorings in codebases. It offers measurable quality‑of‑life improvements, is fully open source, and is just one sudo pacman --sync opencode away. I’m enjoying it so far, and I think it’s the most convenient way to manipulate code with the help of LLMs.