WatchThis

What the company appears to be: WatchThis is a self-reported Windows desktop application that claims to turn one screen recording into a reusable automation skill using AI. The author describes it as an open-source tool built for personal use, with support for multimodal LLMs and integration with platforms like Codex.

What changed: The project was submitted to the OpenAI 2026 hackathon on Devpost, indicating a public launch or demonstration phase. It is described as a personal development effort by one individual (Zhengyang Zhang), with no evidence of prior traction or commercialization.

Single most important open question: Is there any evidence of actual user adoption, revenue, or real-world usage beyond the author’s own description?

The description states that WatchThis is a Windows desktop app built using Electron, React, TypeScript, and Node.js. It captures screen recordings along with system telemetry such as mouse events, keystrokes, clipboard activity, and UI automation data.

It claims to process workflows by:

• Recording live tasks or video inputs
• Using overlapping 60-second segments for long sessions
• Sending both visual feed and real-time event logs to an LLM
• Generating structured workflow drafts
• Allowing user confirmation before finalizing output

The tool outputs a "Computer Use Skill" exportable as ZIP or loadable into platforms like Codex.

Evidence: Self-reported. No independent verification of functionality, performance, or actual use cases.

The author positions WatchThis as:

• A tool that learns from one demonstration ("learn after seeing it just once")
• An open-source and highly versatile automation solution
• Designed to reduce reliance on manual AI input and improve accuracy over time

It is described as an evolution of existing tools that are "lacking in usability" and where AI often misses details.

Inference: The positioning reflects a desire to solve inefficiencies in current workflow automation tools, particularly around precision and ease-of-use. However, this is based on the author’s own claims without external validation.

The description does not explicitly define target customers or ideal customer profiles (ICP). It implies that users might be:

• Individuals looking to automate repetitive computer tasks
• Developers or power users who want reusable automation skills
• Those working with Excel, spreadsheets, or web data entry

There is no indication of enterprise adoption, B2B targeting, or specific verticals mentioned.

Evidence: Not evidenced. The author does not describe any defined customer segments or personas.

No business model or pricing information is provided in the description.

The project is described as:

• Open-source
• Built for personal use
• Intended to be a demonstration tool for a hackathon

There is no mention of monetization, licensing, subscriptions, or paid features.

Evidence: Not evidenced. No indication of how the product would generate revenue.

Technical architecture includes:

• Electron frontend with React and TypeScript
• Native C#/.NET 9 background service for system telemetry logging
• Chrome/Edge browser extension for web context
• Support for multiple LLM backends (Google Gemini, OpenAI-compatible APIs)
• Schema validation of AI responses
• Multi-language localization (5 languages)

Key engineering challenges addressed:

• Aligning video with OS events
• Managing long sessions without loss of context
• Supporting various AI providers dynamically

Development was supported by Codex and GPT 5.6 Sol with ultra mode.

Evidence: Self-reported. No independent confirmation of technical claims or delivery quality.

The author states:

• The project is open-sourced on GitHub
• It has 165 automated tests and packaging smoke tests
• It supports one-click Codex setup and portable ZIP exports
• It was submitted to the OpenAI 2026 hackathon

There is no mention of:

• Users, customers, or adoption metrics
• Revenue or monetization attempts
• Product-market fit validation
• Any form of market traction beyond the author’s own development

Evidence: Not evidenced. No signs of real-world usage or business traction.

The description does not reference competitors or existing solutions in the workflow automation space.

It implies that current tools lack usability and accuracy, but does not compare WatchThis to specific alternatives.

Evidence: Not evidenced. No competitive analysis or positioning against known players.

• Single-person team: Only one developer (Zhengyang Zhang) is listed.
• No revenue or traction data: The product appears to be in early development or demonstration phase.
• Unverified claims: All technical and functional assertions are self-reported with no external validation.
• Limited scope: Designed primarily for Windows desktop use, with no indication of cross-platform support or broader applicability.
• Hackathon project: Submitted to a hackathon suggests it may be experimental or exploratory rather than production-ready.

Inference: These factors suggest limited commercial viability or scalability without further evidence of traction or funding.

1. What specific workflows have you tested WatchThis on, and what were the results?
2. Have you received feedback from other users beyond yourself?
3. How do you plan to monetize this tool if at all?
4. What is your roadmap for expanding beyond Windows desktop and browser extensions?
5. Are there any known limitations or edge cases in how the system handles complex tasks?
6. Can you provide examples of actual skills generated by WatchThis?
7. Do you have plans to support enterprise use cases or integrations with other automation platforms?