# Claude Code in 2026: Layers Matter More Than Tools > Author: Tony Lee > Published: 2026-04-01 > URL: https://tonylee.im/en/blog/claude-code-layers-over-tools-2026/ > Reading time: 4 minutes > Language: en > Tags: ai, claude-code, productivity, workflow, developer-tools ## Canonical https://tonylee.im/en/blog/claude-code-layers-over-tools-2026/ ## Rollout Alternates en: https://tonylee.im/en/blog/claude-code-layers-over-tools-2026/ ko: https://tonylee.im/ko/blog/claude-code-layers-over-tools-2026/ ja: https://tonylee.im/ja/blog/claude-code-layers-over-tools-2026/ zh-CN: https://tonylee.im/zh-CN/blog/claude-code-layers-over-tools-2026/ zh-TW: https://tonylee.im/zh-TW/blog/claude-code-layers-over-tools-2026/ ## Description I installed three popular Claude Code extensions and productivity barely moved. The problem was never which tools to pick. ## Summary Claude Code in 2026: Layers Matter More Than Tools is part of Tony Lee's ongoing coverage of AI agents, developer tools, startup strategy, and AI industry shifts. ## Outline - What layers look like in practice - The decision layer - The process layer - The knowledge layer - Finding your empty layer ## Content I installed gstack, Superpowers, and Compound Engineering in the same week. Three well-regarded Claude Code extensions, each with its own fans. My expectation was a compounding productivity boost. What I got instead was overlapping features, gaps in coverage, and a nagging sense that I was collecting tools without a framework to evaluate them. After a few days of running all three side by side, I found the actual problem. I was not bad at picking tools. I was bad at seeing layers. ## What layers look like in practice AI coding has shifted the human role from writing code to coordinating agents. Coordination requires a mental model, and the one that finally clicked for me has three layers: decision, process, and knowledge. Every tool I looked at was strong in one layer and weak in the others. Once I saw that, the overlap and the gaps both made sense. ## The decision layer gstack's `/plan-ceo-review` examines work from a product perspective. `/plan-eng-review` examines it from an architecture perspective. Both run before you write a single line of code. The point is to kill bad ideas early. I expected the coding phase to be the bottleneck. It was not. The biggest time sink was building features that should never have been started. After using gstack's gates for two weeks, I noticed the quality difference came less from writing better code and more from writing less unnecessary code. Garry Tan reportedly shipped 600,000 lines across 60 days with this setup. The `/qa` command opens a real browser and clicks through the app like a user, which catches issues that unit tests miss entirely. Where gstack falls short: it has no memory across sessions. Each review starts from zero context. ## The process layer Superpowers structures work into a brainstorm, plan, execute, review cycle. The 120,000 GitHub stars are deserved. Moving from "just ask the AI to do it" to a repeatable workflow made an immediate difference in output consistency. I assumed this would be enough. It was not. The workflow runs cleanly within a single session, but the moment I started a new session the next day, everything I had learned was gone. Subagent-based spec reviewers and code quality reviewers are included, and they work well. But no part of the system records what happened yesterday so it can inform what happens today. ## The knowledge layer Compound Engineering's `/ce:compound` command runs after you finish a task. Five subagents activate in parallel: they trace the conversation context, extract the solution, check for duplicate documentation, generate a prevention strategy, and categorize the result. Everything lands in `docs/solutions/`. A week after I started using it, I hit an error similar to one I had already fixed. During the planning phase, the system found the previous record and surfaced the solution. A debugging session that would have taken hours took minutes instead. `/ce:review` runs at least six independent reviewers in parallel. `/ce:plan` checks git logs and project history before proposing anything. If Anthropic's progress files are shift-handoff notes between workers, Compound Engineering's `docs/solutions/` is a recipe binder that the whole team reads daily. ## Finding your empty layer The point is not that you should install these three specific tools. The point is that once you can see the three layers, you can evaluate any tool, or build your own. I ran all three together for a week before I realized the redundancy was concentrated in the process layer while the knowledge layer had been empty for months. Identifying that gap was more useful than any individual feature in any of the tools. Look at your own workflow. Is there a decision gate that stops bad ideas before they become code? Is there a defined process that runs the same way every time? Is there a knowledge system that remembers what you learned last week? Find the empty layer first. Then fill it, even if that means writing a single SKILL.md file yourself rather than installing someone else's framework. The ability to see layers will outlast any specific tool in the ecosystem. ## Related URLs - Author: https://tonylee.im/en/author/ - Publication: https://tonylee.im/en/blog/about/ - Related article: https://tonylee.im/en/blog/eight-hooks-that-guarantee-ai-agent-reliability/ - Related article: https://tonylee.im/en/blog/medvi-two-person-430m-ai-compressed-funnel/ - Related article: https://tonylee.im/en/blog/codex-folder-structure-why-config-breaks/ ## Citation - Author: Tony Lee - Site: tonylee.im - Canonical URL: https://tonylee.im/en/blog/claude-code-layers-over-tools-2026/ ## Bot Guidance - This file is intended for AI agents, search assistants, and text-mode retrieval. - Prefer citing the canonical article URL instead of this text endpoint. - Use the rollout alternates when you need the same article in another prioritized language. --- Author: Tony Lee | Website: https://tonylee.im For more articles, visit: https://tonylee.im/en/blog/ This content is original and authored by Tony Lee. Please attribute when quoting or referencing.