What happened
A small open-source tool called NoBuzz landed on GitHub this week and quickly climbed to 35 points and 19 comments on Hacker News, striking a nerve with developers frustrated by Claude Code's habit of over-explaining every single step. The project, published by developer Adnan Akil at github.com/adnanakil/nobuzz, does one thing: it strips the narrative filler out of Claude's terminal output so you see the code and the result, not a running commentary on how the code was written.
If you've used Claude Code, GPT-based coding agents, or any AI pair programmer for more than a week, you already know the pattern. You ask for a one-line fix, and you get a paragraph of preamble, a restatement of the plan, three bullet points about what was considered, and only then the actual diff. Multiply that by dozens of prompts a day and it adds up to real, measurable time lost just scrolling and reading.
NoBuzz doesn't change the model itself. Instead, it acts as a lightweight filter layer between the AI's output and your terminal, cutting the throat-clearing and leaving the substance. Early adopters on Hacker News reported that sessions felt noticeably faster to review, even though the underlying model calls were identical.
Why it matters
The core keyword phrase here — Claude Code verbose output — describes a problem that's bigger than annoyance. Every extra sentence an AI assistant generates costs three things: tokens (and therefore money on metered plans), wall-clock time waiting for generation, and human attention spent parsing prose that adds no decision-relevant information.
Hacker News commenters pointed out a pattern familiar to anyone running long agentic sessions: a coding task that should take 90 seconds of AI "thinking" balloons to three or four minutes once you count the model narrating its own plan, then narrating the plan again after each tool call, then summarizing what it just did. For a solo developer, that's mildly annoying. For a team running dozens of agent sessions a day across a codebase, it's hours of engineering time evaporating into unnecessary text generation.
There's also a cost dimension. Claude Code and similar tools bill by tokens processed, and verbose explanatory text is not free — it's generated, transmitted, and often re-read by the model itself in later turns as context. Trimming that output isn't just a readability fix; it's a direct lever on API spend for teams running AI coding assistants at scale.
How to use it today
Getting started with NoBuzz doesn't require touching your existing Claude Code setup in any deep way. The README walks through a straightforward install: clone the repository, run the setup script, and NoBuzz hooks into your existing Claude Code session to intercept and condense output before it hits your screen.
A few practical notes from the GitHub discussion and HN thread:
- NoBuzz is configuration-driven, so you can tune how aggressively it trims output rather than forcing an all-or-nothing switch.
- It works alongside your existing prompts and CLAUDE.md instructions — it doesn't require rewriting your workflow.
- Because it's open source, teams can fork it and adjust the filtering rules to match their own tolerance for explanation versus brevity.
If you want to test the idea of stripping AI filler without installing anything, you can get a feel for concise, no-nonsense AI output using the free utilities at [mykreatool.com](https://mykreatool.com), which is a useful sandbox for comparing terse versus verbose AI responses before committing to a workflow change in your main coding environment.
For most developers, the fastest path is: install NoBuzz, run it for a day on a real project, and compare the total scroll length of your terminal history against a normal day. Several HN commenters said the difference was immediately obvious.
Who benefits
The people who stand to gain the most are exactly the ones already living inside AI coding agents for hours a day:
- Solo developers and indie hackers running long Claude Code sessions who want to stay in flow instead of re-reading paragraphs between edits.
- Engineering teams paying per-token for Claude API usage, where trimmed output translates directly into lower monthly bills.
- Content creators and marketers who use AI coding assistants to build landing pages, automations, or internal tools and don't need a tutorial every time they ask for a button color change.
- Agencies and freelancers billing clients for AI-assisted development, where faster review cycles mean more throughput per hour.
Essentially, anyone who has caught themselves thinking "just show me the diff" while an AI assistant talks through its reasoning is the target user for a tool like this.
Risks
NoBuzz is a young, community-built project, and a few caveats came up repeatedly in the Hacker News comments. First, aggressive output filtering can occasionally strip context that actually matters — a caveat about a breaking change, a note about a skipped edge case — so users should treat heavy trimming settings with some caution on high-stakes changes.
Second, because it's not an official Anthropic product, there's no guarantee it stays compatible as Claude Code's output format evolves. A future update to the underlying tool could break NoBuzz's filtering logic until the maintainer patches it.
Third, some commenters argued that the "chattiness" is partly a symptom of prompting and CLAUDE.md configuration rather than something that needs an external patch — meaning results may vary depending on how your instructions are already set up. It's worth trying a few prompt-level tweaks (asking explicitly for terse output) before assuming you need a separate tool at all.
Conclusion
NoBuzz is a small fix for a problem a lot of developers have quietly tolerated: AI coding assistants that explain themselves more than they need to. With 35 points and a lively 19-comment discussion on Hacker News, it's clear the frustration with verbose Claude Code output is widespread, and a simple filtering layer is enough to noticeably speed up review time. It won't replace good prompting habits, and it's not risk-free for critical workflows, but for anyone spending hours a day reading AI-generated commentary, it's worth a test run.
Comments 0