18 entries
Under a budget-capped subscription like opencode Go, picking a model isn't about token price but requests per 5 hours, the one quality source still alive (vals.ai), and tool-calling reliability — the decisive criterion no benchmark measures. The result: one model per preset, a frontier delegated outside the agent, and tools constrained rather than instructed.
An agentic workflow isn't about "using AI to code faster" — it's a restructuring of how you break down, delegate, and verify work. The real gains come from three practices: decomposing tasks into verifiable units, actively managing context, and treating the agent like a competent junior who needs precise instructions.
OpenCode integrates naturally into a Neovim/tmux workflow: a dedicated tmux split, the agent in one pane, your editor in the other. The client/server architecture lets you start OpenCode server-side and connect via SSH with no extra configuration. Full setup in 30 minutes.
Route mechanical tasks to budget or free models (MiniMax 2.5 via Kilo Gateway, Gemini Flash), reserve premium models for reasoning. Result: mechanical tasks (~65% of volume) cost nothing, with no perceived quality loss.
A good coding agent prompt has three components: what you want (precise task), how to validate it (verifiable exit condition), and what you don't want (explicit constraints). Most disappointing results come from instructions missing one of the three.
The context window is a finite resource that degrades. The key is treating it as such: short sessions with verifiable deliverables, persistent context files between sessions, splitting into specialized agents for parallel tasks. Multi-agent workflows aren't added complexity — they're the way to work at scale without losing quality.
The bigger the context window, the later context rot hits — it doesn't go away. The real solution is structuring your work into short sessions with verifiable deliverables. It's an organizational discipline, not a hardware problem to solve.
Kilo Code CLI is a fork of OpenCode with Kilo Gateway on top. In practice, the TUI is identical, custom commands work the same, and migration took 15 minutes. The real difference: free models via the gateway change the workflow economics.
Across 128 auto-scored Claude Code sessions, 68 had a "wrong_approach" friction — 53%. The agent's first instinct is wrong half the time. This isn't a problem to fix, it's a pattern to integrate into the workflow.
Claude Code is an excellent starting point — zero config, immediate quality. But Anthropic lock-in prevents optimizing costs per task, and restricting the Claude Code subscription in third-party tools forced the migration. OpenCode then Kilo Code CLI (its fork with free gateway) solve both without sacrificing quality.
Three patterns that naturally emerged in a daily agentic workflow: automated morning briefing, versioned custom commands in Git, and parallel sessions in tmux splits. Each came from a real friction, not from theory.