AI Coding Assistants Lose Focus Mid-Project — This Tool Fixed It. GSD Hits 32K GitHub Stars
GSD (Get Shit Done), an open-source tool that solves the 'context rot' problem — where AI coding assistants degrade in quality during long sessions — has surpassed 32,000 GitHub stars. Install it with a single command and Claude Code will systematically plan, verify, and auto-commit your work.
Have you ever handed a large project to an AI coding assistant, only to watch the code quality plummet after a while? This phenomenon is called 'Context Rot'. As the AI works, the conversation window accumulates so much information that it's like having a desk buried under a mountain of paperwork — you can't find the important documents anymore.
GSD (Get Shit Done) is an open-source tool that tackles this problem head-on. It has surpassed 31,900 GitHub stars, and engineers at Amazon, Google, and Shopify are using it in production. It also hit the Hacker News front page today, sparking a lively discussion.
Why AI Loses Focus During Long Sessions
AI coding assistants like Claude Code have a fixed amount of information they can process at once (called a context window — think of it as the AI's short-term memory). As a project grows, this space fills up with previous conversations, code snippets, and error messages, causing the AI's judgment to deteriorate. Research from Carnegie Mellon University even found that using AI coding tools for more than two months can actually slow developers down.
GSD solves this with a 'fresh brain for every task' approach. It breaks large projects into small units and opens a clean new conversation window (200K tokens) for each task. This way, the AI stays sharp no matter how large the project gets.
6 Steps That Turn AI Into a Well-Organized Dev Team
The core idea behind GSD is making the AI cycle through the roles of Product Manager → Researcher → Architect → Developer → Tester — even if you're coding solo.
GSD Workflow: 6 Stages
The AI asks questions to fully understand your requirements. It clarifies what you're building, who will use it, and what features are essential — then auto-generates a project brief.
Ambiguous details are resolved upfront. For a UI, that means decisions like button placement and empty-state handling. For an API, it covers error formats and response structures.
The AI researches relevant technologies and creates an execution plan broken into 2–3 task groups. Each plan clearly states what to build and how to verify it.
Independent tasks run in parallel simultaneously. Each completed task automatically gets a Git commit, so if something breaks, you can roll back just that specific task.
The finished output is reviewed item by item. If issues are found, the AI automatically diagnoses the cause, creates a fix plan, and re-executes.
The process moves to the next phase and repeats. Once all phases are complete, a release tag is created and work begins on the next version.
'Wave Execution' — Run Non-Overlapping Tasks Simultaneously
The standout feature of GSD's execution stage is wave-based parallel execution. For example, if you're building an e-commerce app:
WAVE 1 (parallel): User Model + Product Model
WAVE 2 (parallel): Order API (needs User Model) + Cart API (needs Product Model)
WAVE 3: Checkout Screen (needs Order + Cart APIs)Independent tasks run simultaneously, while tasks that depend on previous results execute in order. Since a Git commit is automatically created after each task, you can pinpoint exactly which task introduced a bug later on.
What Real Users Are Saying
Reactions in the Hacker News discussion were mixed.
👍 Positive Feedback
User yoaviram said, "I've been using it for three months and complex tasks come out at 95% completion quality," adding that they shipped a SaaS product with GSD. prakashrj shared, "I'm a systems engineer with zero app development experience, but I used GSD to generate 250,000 lines of code in one month and built an app."
👎 Negative Feedback
On the other hand, yolonir criticized it as "overly complex — it burns through API usage quickly," and testycool pointed out that "the 5-hour session limit was reached in just 30 minutes," highlighting the heavy token consumption.
The central debate is whether a structured framework delivers real productivity gains relative to its token consumption. The general consensus: it's highly effective for large-scale projects, but can be overkill for simple tasks.
Install and Get Started in 5 Minutes
GSD supports all major AI coding tools including Claude Code, Gemini CLI, GitHub Copilot, and OpenCode. Installation takes a single command.
# Install (interactive — lets you choose which tools to set up)
npx get-shit-done-cc@latest
# Install directly for Claude Code only
npx get-shit-done-cc --claude --global
# Install for all AI coding tools at once
npx get-shit-done-cc --all --globalAfter installation, type /gsd:help in Claude Code to see available commands. To start a new project:
# Start a new project — the AI asks questions and creates a project brief
/gsd:new-project
# Quick task — run a simple task without the full workflow
/gsd:quick "Build a login page"
# Check current progress
/gsd:progressHow It Compares to Superpowers and gstack
AI coding workflow tools have been popping up everywhere recently. Here's how they compare:
GSD's key differentiator is that it opens a fresh context for every task, preventing quality degradation at the source. The trade-off is higher token consumption, so the practical approach is to use regular Claude Code for simple fixes and GSD for large-scale projects.
Who Should Use This
If you're a solo developer looking to build an app on your own, GSD systematically manages the entire process from planning to verification. Even with minimal coding experience, you can tell the AI "build me this app" and GSD will ask the right questions, plan, and execute on your behalf.
If you're a developer frustrated by inconsistent quality from AI coding tools, GSD's structured workflow helps maintain consistent output. Just keep in mind that API costs may increase, so weigh it against the scale of your project.
What's Next
GSD has been updating rapidly — from v1.22 to v1.25 in March alone. Recent additions include Google's Antigravity runtime support and natural language command routing (/gsd:do).
AI coding tools are evolving beyond "writing code for you" to "managing your entire project." As tools like GSD, Superpowers, and gstack compete, 'vibe coding' is moving past a passing trend and establishing itself as a structured development methodology.
Related Content — Get Started with AI Using EasyCLCo | Free Learning Guide | More AI News
Stay updated on AI news
Simple explanations of the latest AI developments