--- Need a "Canvas"-like feature for direct interaction with agent outputs or a virtual environment where the agent can edit autonomously: Reason: In fact, this is the most emphasized feature of the latest model, Claude-Opus-4.6, yet it is not being utilized properly here. Furthermore, although the "Project" feature has been around for a long time, it effectively serves only as a chat archive. Need a feature to regenerate or edit outputs: Reason: Due to the nature of language models, if even a single incorrect output occurs, additional corrections are required. This is a critical issue because even if corrections are made, the subsequent flow of the conversation can change. It would be better to separate the code execution function and the search function within "Enable advanced research & reasoning": Reason: There are situations where only code-based calculations are needed, and others where only research-based internet searches are required. Therefore, it seems necessary to separate these options. This is somewhat similar to point #1, but it would be great if the agent could output created projects or files in the form of shareable links: Reason: Currently, the formats of outputs produced by the agent are very limited, existing only as text and charts (when using Compute). I wish to be able to manipulate detailed options for the Auto Agent: Reason: The Auto Agent has the ability to automatically route models and execute the "Enable advanced research & reasoning" feature. However, it is opaque as to which model the Auto Agent has been routed to, making verification impossible. Additionally, since it triggers "Enable advanced research & reasoning" dynamically, the boundaries for triggering are ambiguous, which actually discourages its use. I hope detailed options to control the Auto Agent will be implemented to fit its nature as a dynamic agent. ARI Individual Credit Purchase Feature: Reason: As far as I know, the only ways to currently use ARI are by subscribing to the Max plan or contacting Enterprise sales. However, both methods involve high initial costs, creating a high barrier to entry. Since there are non-power users who occasionally need this feature, I hope you will sell credits that allow users to utilize ARI on an as-needed basis. Reasoning Depth Adjustment Feature: Reason: Latest models like Gemini, Claude, and GPT allow for manual adjustment of reasoning depth. I hope this feature is implemented so users can utilize it appropriately according to their needs. --- Some of these items are suggestions that were considered a long time ago but continue to be delayed. It has been almost over a year since those suggestions were first raised.
