AI Coding Notes 1 | /home/rook1e

Outsource the grunt work of CRUD to a Coding Agent, freeing up your energy for discussing solutions, breaking down tasks, reviewing code, and tackling hard problems. Enjoy the pure joy of creation.
The loom of software development has arrived. AI Agents won’t fully replace programmers, and probably won’t reduce the number of jobs in the long run either, but they will reshape how the entire industry operates. We may see software engineering practices, programming languages, and system architectures better suited for AI Agents, driving digitalization across all aspects of society at greater scale, lower cost, and higher efficiency.
Burnout comes easier. AI thinks and generates far faster than human eyes can read and minds can process. Auditing every line of AI-generated code and context-switching between multiple task sessions is mentally exhausting.
Claude Code + Opus/Sonnet 4.5 is still SOTA, but the moat isn’t that deep.
GLM 4.6 is an excellent workhorse model: fast, cheap, and generates at about 80% quality — perfect as an executor after you’ve designed the approach/plan with a SOTA model.
OpenCode is a solid open-source alternative to Claude Code. OpenCode has many strengths, such as built-in LSP, easy model switching, and a built-in HTTP API. But the downsides are equally significant — it’s not stable enough, and the system prompts still need polish. Promising for the future.
Stick with mainstream tech stacks.
Don’t buy annual subscriptions for AI products.