0xSero interviews Vaibhav Srivastav, known as VB, about GPT-5.6 and the product layers that turn models into useful agents. Vaibhav Srivastav describes different model options and reasoning modes, then explains the effort to make tools, browser interaction and longer tasks easier to use across Codex and ChatGPT. The episode captures the participants' account at the time of recording rather than a guarantee of current feature availability.
0xSero and Vaibhav Srivastav exchange examples of agents helping with travel planning, administrative research and organizing information across connected services. Vaibhav Srivastav argues that onboarding should start with an accessible first task, then progressively connect the tools a person already uses. He also discusses asking clarifying questions during work and using feedback to distinguish isolated problems from issues affecting many users.
0xSero points to failures in benchmark configuration and unrealistic time estimates as examples of capability gaps that can waste resources. Vaibhav Srivastav identifies learning from mistakes, keeping useful task history, supporting long-running threads and choosing when follow-up loops are appropriate as areas he would like to improve. These closing remarks are personal aspirations, not a confirmed research plan or a promise that autonomous follow-ups are always appropriate.
Watch on YouTube
