--- title: "Claude Code /powerup and /insights: Fixing the 80% Problem" date: 2026-04-12 tags: ["claude-code","anthropic","developer-experience","onboarding","productivity"] categories: ["AI Tools","Guides"] summary: "Most developers use a fraction of what Claude Code can do. Two new commands shipped in v2.1.90 — /powerup and /insights — attack this problem from opposite ends: one teaches you what's possible, the other shows you where your actual workflow breaks down." --- Claude Code has a discovery problem. The tool is remarkably deep — hooks, sub-agents, custom slash commands, CLAUDE.md memory, /ultraplan, skills, the full MCP ecosystem — but nothing in the default experience tells you any of it exists. You open the terminal, ask Claude to write some code, and it obliges. The other 80% of what the tool can do sits undiscovered. Anthropic shipped two commands in v2.1.90 (April 1, 2026) that attack this problem from opposite ends. `/powerup` teaches you what's possible; `/insights` shows you where your existing workflow is leaking efficiency. Together they form something that looks a lot like Anthropic's acknowledgment that "just ship powerful features" isn't a complete product strategy. ## /powerup: Duolingo for Your Terminal The premise is simple: 18 interactive, animated lessons built directly into the CLI, no browser required. You run `/powerup`, select a lesson with arrow keys, and Claude Code walks you through the feature hands-on. The lessons follow a consistent four-part structure: 1. **Concept introduction** — what this feature is, in plain language 2. **Guided exercise** — you try it, in your terminal, in your actual project 3. **Feedback** — Claude Code responds to what you did 4. **Progression hook** — context for why the next lesson builds on this one The full curriculum runs from fundamentals to intermediate automation: **Beginner tier**: basic context management, `/clear` vs. `/compact`, the CLAUDE.md memory system, plan mode (Shift+Tab), model selection. These are the features most new users discover by accident after weeks of daily use — if at all. **Intermediate tier**: the skills and custom commands system, hooks (`PreToolUse`, `PostToolUse`), sub-agent orchestration, MCP server configuration, `/rewind` and checkpointing. The time investment per lesson is roughly 10 minutes. Working through the full curriculum in focused sessions would take under three hours — less time than most teams spend misconfiguring MCP servers by trial and error. ## Context-Aware Instruction What elevates `/powerup` past a static tutorial is that it reads your project state before presenting each lesson. The CLAUDE.md module checks whether you already have one and adjusts accordingly — if you do, it analyzes your existing configuration and suggests improvements rather than walking you through creation from scratch. The context management lesson behaves differently in a 200-file monorepo than in a three-file script. This matters because the gap between "I know this feature exists" and "I know how to apply it to my actual codebase" is where documentation typically fails. Generic examples don't transfer. Context-aware instruction does. ## /insights: Your Workflow Under a Microscope While `/powerup` is forward-looking, `/insights` is retrospective. It analyzes your local Claude Code session history from the past 30 days and generates a full HTML report covering: - **Recurring patterns** — what kinds of tasks you actually give Claude Code - **Friction points** — where you repeatedly retry, correct, or restart - **Tool usage distribution** — which Claude Code capabilities you use and which you never touch - **Workflow inefficiencies** — sequences of actions that suggest a better-configured approach The analysis runs entirely locally. Session logs live in `~/.claude/`; the HTML report is generated and saved there too. Nothing leaves your machine. For teams on enterprise plans with strict data governance requirements, this is the right answer to "can we analyze how developers use AI tools without shipping private data to a third party." The feature that will get the most immediate use is **automatic CLAUDE.md rule generation**. `/insights` identifies instructions you've repeated multiple times across sessions — things you've typed to Claude Code again and again — and converts them into ready-to-paste CLAUDE.md configuration rules. If you've told Claude "don't add console.log statements" four times this month, `/insights` turns that into a permanent instruction. ## The 80% Problem, Stated Plainly JetBrains' January 2026 developer survey found Claude Code at 18% work adoption among professional developers — up 6× year-over-year — with an industry-leading 91% customer satisfaction score. The tool is being adopted. The question is depth of adoption. There's a common pattern in powerful developer tools: new users find the one workflow that solves their immediate problem and stop there. Git users who never learn `rebase -i`. vim users who stay in insert mode. Claude Code users who write code with it but never configure hooks, never build custom skills, never use multi-agent workflows for the tasks that genuinely benefit from them. The cost of shallow adoption isn't zero. A developer who uses `/compact` correctly on long sessions gets substantially better results than one who doesn't. A team that codifies standards in CLAUDE.md gets consistent behavior across agents. An engineer who knows how to hand off to sub-agents can parallelize work that would otherwise be serial. The efficiency delta between a casual user and a configured power user of Claude Code is not marginal. `/powerup` and `/insights` are Anthropic's admission that shipping features isn't enough — you have to guide people to them. That's a product decision, not just an engineering one. ## How to Use Both Effectively The practical approach is to run `/powerup` first, working through the lessons in order over a few sessions, then run `/insights` after another week or two of use to see which features you've actually incorporated into your workflow. The second `/insights` run becomes a feedback loop: did the things you learned from `/powerup` actually change your behavior? For teams, the auto-generated CLAUDE.md rules from `/insights` are worth extracting and codifying into a shared team configuration. Instructions that appear repeatedly across individual developers' session histories are candidates for the team-level `CLAUDE.md` — those are the implicit norms that should be made explicit. ```bash # Learn what you're missing /powerup # After a week of use, analyze your patterns /insights ``` Both commands are available in Claude Code v2.1.90 and later. If you're running an older version, `npm update -g @anthropic-ai/claude-code` will get you there. The discovery problem isn't fully solved by two commands. But these are a real step toward the kind of in-product education that separates tools people use from tools people use well. --- *Sources: [Claude Code v2.1.90 release notes](https://code.claude.com/docs/en/changelog) · [claudefa.st /powerup guide](https://claudefa.st/blog/guide/mechanics/claude-powerup) · [Claude Lab /powerup command guide](https://claudelab.net/en/articles/claude-code/claude-code-powerup-command-guide) · [Claude Lab /insights command guide](https://claudelab.net/en/articles/claude-code/claude-code-insights-command-guide) · [JetBrains developer survey Jan 2026](https://blog.jetbrains.com/research/2026/04/which-ai-coding-tools-do-developers-actually-use-at-work/)*