# What Claude Code Can Do That You Haven't Tried - Gui Ferreira - NDC Oslo 2026

## Executive summary

This advanced session provides a deep dive into maximizing AI coding assistants, specifically Claude Code, moving beyond basic code generation. The focus is on advanced configuration, workflow automation, and treating the AI as a comprehensive assistant for non-coding tasks. Key strategies include managing project context via `Claude.md`, using specialized commands (`/goal`, `/focus`), and implementing robust development practices like automated testing hooks and parallel work streams.

## Key takeaways

- Optimize Project Context with `Claude.md`: Do not over-document `Claude.md`. Studies (ETH Zurich) suggest that efficient, 'lean' configuration files are best. Use the `/doctor` command to audit and clean up redundant information, moving specific rules to language-specific files within `.cloud/rules` to prevent context confusion.
- Implement Structured Workflows and Iteration Control: Use 'Plan Mode' (Shift+Tab) to generate a plan, then open it as a markdown file (`Ctrl+G`) for collaborative editing. To prevent context window bloat during complex refactoring, utilize the `/rewind` feature to step back to previous states, and use the `/focus` command to filter out intermediary messages and see only the initial prompt and final result.
- Automate and Parallelize Development Tasks: Leverage specialized skills for complex tasks: use `/patch` for large-scale refactoring/migrations, `/loop` to monitor deployments (e.g., checking GitHub Actions status), and `/goal` to define a task as a verifiable journey (e.g., 'Test passes' or 'PDF is generated').
- Manage Context and Cost Proactively: Monitor resource consumption by running `/context` in an empty session to diagnose unnecessary data polluting the context window. When debugging, proactively run `/compact` to summarize history and prevent the 'danger zone' (around 40% context usage) where token efficiency drops.

## Technical details

- Configuration & Context Management: Use `slash import [product name]` to migrate project configurations (MCP servers, sub-agents) from other coding agents. Store user-specific, session-wide preferences in the user profile folder (`.cloud/`) rather than repeating them in every project's `Claude.md`.
- Workflow Commands: • `/doctor`: Audits and cleans up `Claude.md` by identifying duplication and suggesting moves to skills. • `/goal`: Defines a task as a verifiable journey with specific success criteria (e.g., 'Test passes'). • `/focus`: Filters conversation logs to show only the initial prompt and the final result, simplifying long processes.
- Advanced Interaction & Debugging: • `Ctrl+G`: Opens the plan generated by Claude as an editable markdown file in the editor. • `!`: Prefixing a command (e.g., `!npm test`) forces the AI to treat the following input as a raw bash command, bypassing unnecessary reasoning. • `slash BTW`: Runs a query as a side task that does not pollute the main conversation's context window.
- Parallel Development & State Management: Use 'work trees' (a Git concept) to work on multiple features (e.g., bug fix, refactor, new feature) simultaneously on the same project without committing. Use `slash color` and `slash rename` to maintain visual context when switching between parallel tasks.

## Practical implications

- Adopt a proactive approach to context management by regularly running `/compact` and auditing `Claude.md` to ensure the AI is working with the most efficient, non-redundant information.
- Treat the AI assistant not just as a coder, but as a project manager and documentation tool. Use it to summarize meeting notes, organize file structures, or generate quarterly review summaries.
- For build engineers, utilize the 'Plan Mode' and 'Goal' features to formalize complex, multi-step processes (like migrations or deployments) into verifiable, iterative tasks, reducing manual oversight.
- Be mindful of token consumption. Track costs using the terminal status to understand the true cost of complex features, especially when running multiple sub-agents.

## Topics

AI Coding Assistants, Context Management, Workflow Automation, Build Engineering Practices, Generative AI, DevOps/CI/CD, Anthropic Claude Code, ETH Zurich Study, Obsidian

Source: https://www.youtube.com/watch?v=P3TCZ2xrvbo