REVIEWS / AI CODING / OWNER INSIGHTS
🦉 WE READ 210 OWNER COMMENTS
Prime Agent: what owners actually say
Owners love the concept of AI-driven web modification but hit execution failures and serious privacy concerns
What owners complain about
- Privacy and data handling COMMON
Multiple users flagged that the API service receives browser data and there is no clear privacy policy for it. Concerns specifically include the agent potentially capturing passwords, account numbers, and other sensitive input. One user noted the privacy policy says data is shared with the API service and questioned whether do-not-train/retain provisions are trustworthy.
- Tweak execution failures SOME
A user tested increasing YouTube information density and it failed: took 180+ seconds, only shrank some thumbnails while leaving padding intact so no real density was gained. The developer acknowledged it as a fail.
- No local model support SOME
The agent cannot run locally. Developers tested many open/closed models and could not get open models to perform to their standards, meaning all processing goes through cloud APIs.
- Redundant with existing tools SOME
Several users pointed out that LLMs already generate Violentmonkey/Tampermonkey scripts with minimal prompting, and questioned why a VC-backed extension is needed when open-source tools and community blocklists already exist.
- No clear monetization or moat SOME
Users noted the product has no moat and can be copied, and the developers themselves admitted uncertainty about how they will monetize, calling it dependent on cloud credits and investment.
What owners love
- YouTube Shorts removal one-shot success
Removing all Shorts sections and buttons from YouTube.com worked on the first attempt with no apparent issues.
- Fills a real gap
Users see it as a needed step toward fixing the web, citing concrete pain points like GitHub code preview being broken (symbols not showing, inability to view code properly).
- Persistent memory and learning
The Factorio learning environment demo showed Prime Agent can turn things learned during a run into persistent memories and skills, which excited users.
- Developer responsiveness
The developer actively engaged with HN commenters, acknowledged failures honestly, and iterated on feedback in real time, which users appreciated.
Surprising patterns
- Users immediately saw a B2B angle: approaching websites with the modifications users are making and selling that as product insight back to site owners.
- Some users questioned whether pages in the future should fundamentally be designed differently if tools like this become common, seeing it as a signal of broken web design.
- A user discovered that even after a failed density tweak, clicking 'New Modification' to re-select the updated version partially fixed the issue, suggesting iterative refinement is built into the workflow.
WHO SHOULD SKIP IT
Privacy-conscious power users who already run Tampermonkey/Violentmonkey with LLM-generated scripts and are uncomfortable routing browser data through a cloud API with no local model option.
Synthesised from 210 real owner comments across 6 platforms. Every point is grounded in the comments — no marketing, no AI guessing. How we do it →