[OpenCode Field Notes Part 1] Why did work get harder even with more tools?

1. Why did speed drop when tools increased?

There are many LLM-based vibe-coding tools now.

When ChatGPT first appeared, I spent too much time in the browser doing repetitive copy/paste work. Then I moved heavily to agent-style tools like Cursor. At the time, auto mode was slow but effectively unlimited, so the subscription felt worth it. In urgent periods, I even paid extra token costs each month.

The turning point was policy change. After unlimited auto mode disappeared, I became more interested in CLI tools like Google Antigravity and Claude Code.

At first, editing in CLI felt unfamiliar, so I used embedded terminals in VS Code/Zed. Later I moved to dedicated terminal workflows. Eventually I built a routine: first-pass work on phone + tmux remotely, then detailed edits on laptop.

But one core pain remained. As sessions grew longer, old context started hurting current work. Responses slowed down, and manually handing off context to a fresh session became exhausting.

In short, long sessions kept accumulating tokens and context overhead, so I needed a better approach.

2. Core problems I faced

• Re-setup cost was high because each tool had different configuration styles
• Prompt/command patterns were inconsistent across tools
• Token/subscription limits interrupted work mid-flow
• In long sessions, bloated context reduced both speed and accuracy

For someone using LLMs all day, these issues were critical. I was spending more time managing tool constraints than solving actual problems.

3. My selection criteria

I started evaluating tools by these criteria first:

1. Can I sustain the same routine for long sessions?
2. Can I operate models/accounts flexibly?
3. Can I adapt it to my style via plugins/agents?

By this standard, OpenCode fit best for me.

4. Next

In the next post, I will cover my 10-minute onboarding flow: installation, first run, and initial setup order.