Matt Van Horn Shares Claude Code Workflow Development Priorities
Matt Van Horn published a lengthy article detailing all the tips and workflows for using Claude Code, advocating for the abandonment of traditional IDEs in favor of relying solely on plan.md files and voice for development.
He uses /ce:plan as the first response command, triggering multiple research agents to analyze the codebase, historical documents, and best practices in parallel, generating a structured plan.md that includes problem definitions, solutions, a list of file changes, and acceptance criteria with checkboxes. The /ce:work command then executes the plan, breaking down tasks, writing code, and validating. In the past 30 days, he has submitted 263 commits, accumulating 70 plan files, and has become the third-largest contributor to the Compound Engineering plugin on GitHub.
This workflow has rapidly spread within the developer community, with the Compound Engineering plugin (install command /plugin marketplace add EveryInc/compound-engineering-plugin) solidifying the plan-driven model into a mandatory process, promoting AI assistance from merely aiding coding to full-process engineering compounding.
Source: Public Information
ABAB AI Insight
Matt Van Horn has previously developed several Claude Code plugins, including last30days-skill, and has continuously submitted 21 commits to Compound Engineering, becoming a core contributor. He has also publicly demonstrated how to use a similar agent system to help non-engineers quickly build products, reflecting his long-term investment in the AI coding toolchain.
In terms of capital and resource pathways, Every Inc. promotes Compound Engineering as an open-source plugin, quickly attracting developer contributions and accumulating project-specific knowledge (AGENTS.md / plan.md) through marketplace mechanisms and GitHub collaboration, forming a positive feedback loop that reduces the marginal cost of subsequent tasks while accumulating user stickiness and data assets for the Every ecosystem.
Similar cases include the evolution of AI coding tools like Cursor and GitHub Copilot, as well as the early transformation of Agile methodologies into tooling; Matt's practice is currently at a transitional phase where AI-driven development shifts from experimental stages to mainstream workflow control, emphasizing "each engineering unit makes the next easier" rather than accumulating technical debt.
Essentially, this represents a technological substitution and industrial chain reconstruction: AI agent clusters replace manual planning and context maintenance, reconstructing the developer productivity chain through structured planning documents and historical compounding mechanisms, shifting pricing power from single IDE vendors to the open-source plugin ecosystem and personal knowledge base owners, as high-quality plan.md files become reusable and tradable forms of capital.
ABAB News · Cognitive Laws
Think first, execute later: build a plan before writing code, doubling the success rate for complex tasks.
Each unit of work is not a consumption, but a compound interest: today's accumulated plan.md brings tomorrow's costs close to zero.
Tooling habits surpass genius inspiration: mandatory processes turn "common knowledge" planning into compounding achieved by a few.