Agentic AI Coding: Slash Commands for Context and Cost Control

A founder details a structured workflow using Qwen Code's slash commands to manage context and reduce token costs when building a desktop application. This approach emphasizes disciplined AI interaction.

A founder reports using Qwen Code, an open-source agentic coding CLI, to build Achu, a desktop screenshot beautifier app. The workflow centers on a specific set of slash commands and context management strategies, aiming to reduce token costs and maintain development context across sessions. This approach details a disciplined method for integrating AI into daily coding tasks, moving beyond generic prompts to structured interactions.

Initializing Project Context with /init

The founder begins new projects or returns to existing ones with /init. This command analyzes the directory, generating an initial context file that maps the project's structure and key files. Following this, the founder manually adds descriptive paragraphs about the project, current milestones, tech stack (Electron, React, TypeScript), and known constraints (e.g., Electron IPC boundaries, Upstash Redis, Gumroad monetization). This upfront investment saves enormous amounts of back-and-forth later.

Spec-Driven Planning via /plan

Rather than broad requests, the founder uses /plan to initiate a structured planning mode within Qwen Code. This command is employed for new feature development, such as "Implement the Privacy Guard redaction pipeline." The process involves iterative planning sessions, ensuring a detailed specification is established before any code generation begins. This method emphasizes defining requirements upfront.

Context Hygiene and Cost Management

The workflow prioritizes maintaining context and controlling token expenditure. The founder uses /compress to reduce the active context window, and /clear to reset it when necessary. For persistent architectural decisions or crucial information, /remember is used to store details across sessions. The /btw command allows for quick, out-of-band comments without polluting the main context. To manage costs, a fast model (qwen3-coder-flash) is selected with /model --fast for lighter tasks like /recap or prompt suggestions, while the more powerful Qwen Max handles complex reasoning.

Leveraging Source Code and Subagents

The founder directs Qwen Code to analyze library source code directly, rather than relying on potentially outdated or incomplete documentation. The system's subagent support is also utilized for parallelizing independent development tasks. This strategy aims to provide the AI with the most accurate and granular information, improving the quality of generated code and accelerating development.

What We'd Change

The described workflow relies heavily on the specific features and command structure of Qwen Code, an open-source CLI. While effective for a solo founder building a desktop app, the direct transferability to larger teams or different tech stacks is not guaranteed. The manual addition of project context after /init could become a bottleneck for projects with rapidly evolving architectures or multiple contributors. Automating this initial context capture or integrating it with existing documentation systems would be necessary for enterprise adoption.

The founder reports significant productivity gains and cost reductions, but these claims lack quantitative evidence. Without metrics like lines of code generated per hour, bug reduction rates, or actual token cost savings compared to alternative methods, the efficacy remains anecdotal. Future iterations of this playbook would require instrumenting the development process to validate these benefits with data. Furthermore, relying on pointing the model at library source code, while effective, assumes readily available and well-structured source. This might not hold true for proprietary libraries or complex, undocumented internal systems.

The founder's approach to AI-assisted coding with Qwen Code demonstrates that disciplined interaction, rather than simple prompting, can yield tangible workflow benefits. By treating the AI as a structured collaborator through specific commands and rigorous context management, development can become more efficient. This method emphasizes upfront planning and continuous context hygiene, providing a template for founders seeking to integrate agentic AI tools into their daily build cycles.

The investor read

This signal highlights the growing maturity of agentic AI coding assistants, particularly open-source solutions like Qwen Code, as viable alternatives to commercial offerings for bootstrapped founders. The emphasis on context management and cost optimization reflects a market demand for efficient, controlled AI integration in development workflows. Investors should note the shift towards CLI-based, highly configurable tools that allow developers to maintain control over their environment and data. While Achu itself is a lifestyle product, the underlying trend of AI-driven developer productivity tools, especially those that reduce cloud spend or accelerate time-to-market for solo founders and small teams, represents a significant investment area. The ability to articulate and measure productivity gains will be crucial for ventures in this space.

Pull quote: “This upfront investment saves enormous amounts of back-and-forth later.”