Tech • AI • Robotics • Game
GPT-6 Astra + ChatGPT Work Changes Everything
9.1/10TL;DR
GPT-6 Astra inside ChatGPT Work is positioned as a higher-end task automation setup, with strengths in file analysis, writing personalization, app creation and voice-driven task delegation, but with significantly higher usage costs.
KEY POINTS
Work targets end-to-end tasks
ChatGPT Work is presented as a mode designed to complete full tasks rather than just assist with ideas. Instead of only brainstorming a presentation, it can generate the presentation, save the file, and revise it inside the same workflow. The desktop app is described as the best environment for this setup.
Model and thinking level affect speed, quality and cost
Users can choose GPT-6 Astra and set a thinking level that controls how much effort the model spends before answering. Higher settings are described as more thorough and structured, while lower settings are faster and cheaper. A medium setting is framed as a practical balance, especially because GPT-6 is described as strong even at its lowest setting.
Folder-level access enables broad business analysis
A central feature is the ability to attach an entire computer folder to a project so the model can read and edit files inside it. In one example, a mock plumbing business folder was connected to a new project, giving the model access to records and internal documents. That setup was then used to request a full business audit covering weaknesses, strengths and planning for the next month, quarter and year.
The model generated a report and flagged hidden issues
After reviewing the attached files, ChatGPT Work created a new file directly on the computer called owner planning brief. The report was presented in a visual format with tables rather than only long text. It also detected planted problems, including money routed to an unauthorized beneficiary and overdue customer payments, then turned those findings into an action plan.
Writing can be customized through reusable skills
Another highlighted use case is writing with a reusable skill, effectively a saved instruction set. A Write Like Me skill can be built by attaching two to four authentic writing samples and asking the system to model tone, structure and phrasing. A higher thinking level is recommended during skill creation so the style profile is more accurate, after which the skill can be reused at lower cost.
Users are warned against training on AI-shaped writing
The guidance stresses that writing samples should be genuinely authored by the user. Pulling from email archives or folders with heavy prior AI assistance risks teaching the model to imitate AI-generated prose rather than a distinct personal voice. Once the skill is built, it can be invoked during any task to turn research or notes into text that matches the user’s style.
Non-technical users can build apps and games
The system is also framed as a simple entry point to vibe coding, letting non-programmers describe an app or game in plain language and have the model generate the files. In a demonstration, a folder was created for a Pokémon-like monster-catching game, and the model built a playable native app inside that location. The game included movement, combat, creature capture and healing mechanics.
Computer use helps test and refine generated software
A notable capability is that the model can request permission to open the generated app and test it on the computer. That allows it to identify bugs or usability issues before handing the project over. In the game example, the first version was playable but needed smoother movement and better graphics, showing both the practicality and the current limits of automated software generation.
Live voice can delegate across multiple tasks
Live voice is described as a control layer for ongoing work. Spoken requests can be routed to separate tasks, such as sending gameplay revisions to the game-building task while also asking another task to draft an email about the plumbing audit. In one correction, the system also distinguished the game name Mosswood from the starter creature Mossling, showing contextual memory across tasks.
The main trade-off is price
Despite the performance gains, GPT-6 is described as expensive enough to require selective use. Routine scheduled automations such as inbox checks or trending-news scans are considered better fits for 5.6, where quality differences were noticeable but not meaningful enough to justify the higher cost. The practical recommendation is to run the same task on both models and reserve GPT-6 Astra for cases where the output gap is substantial.
CONCLUSION
The combination of GPT-6 Astra and ChatGPT Work points to a more agent-like form of desktop AI that can analyze files, create documents, build software and coordinate tasks by voice. Its strongest appeal lies in high-value work where better reasoning and computer use outweigh the added cost.
Explain this