AI Summary
→ WHAT IT COVERS Scott Tolinski and Wes Bos break down the full landscape of AI coding tools in 2026, covering models, agents, sub-agents, skills, slash commands, hooks, plugins, and MCP servers — clarifying what each component does, where it lives, and when to actually use it. → KEY INSIGHTS - **Model selection matters significantly:** Codex 5.3 performs better for precise JavaScript logic while Opus 4.6 suits exploratory and creative work. Switching models based on task type produces measurably better results than using one model for everything. Anthropic's new fast mode runs 2.5x faster but costs 6x more, making it impractical for most workflows. - **Context bloat degrades output quality:** Stuffing agents.md with thousands of lines slows every session and muddies results. The file should contain only essential project-wide facts — language, framework, coding conventions — nothing more. Skills solve this by loading context conditionally, only when the AI determines a specific capability is needed for the current task. - **Agent vs. skill decision rule:** Use a skill for one-off tasks where the AI executes and finishes. Use an agent when the workflow requires back-and-forth iteration, auditing, or multi-step refinement. Combining both — an agent that calls a skill — can work but risks over-engineering, adding compute cost and latency without proportional output improvement. - **Slash commands offer underrated precision control:** Unlike skills that the AI invokes autonomously, slash commands function like callable functions with arguments, giving developers direct control over reusable prompts. Practical uses include scaffolding new routes, running linters, and triggering tests. They can be mapped to hardware shortcuts like a Stream Deck for faster invocation. - **Tool-model independence is a strategic priority:** Locking into Claude Code means locking into Anthropic's pricing and model choices. Tools like OpenCode, Py, and Charm Crush allow model switching between Codex, Opus, Grok, and others without changing workflows. As open-source models improve, razor-thin model margins will make tool flexibility the primary competitive differentiator. → NOTABLE MOMENT Wes ran the "superpowers" GitHub skill pack — a full TDD, Git worktree, sub-agent workflow — to build a simple two-page cottage information website. The process ran for three and a half hours and cost $26 in API fees, producing a result achievable in minutes with a basic prompt. 💼 SPONSORS [{"name": "Sentry", "url": "https://sentry.io/syntax"}] 🏷️ AI Coding Tools, LLM Model Selection, Coding Agents, MCP Servers, Developer Workflow