OpenAI gave agents more to do. Teams still need to decide what work to give them.

The rest of the announcements point in the same direction. GPT-6.1 Sol offers a more capable model for coding, computer use, and professional tasks at lower standard token prices than Astra. ChatGPT Space and Pages give teams places to share context and work with AI together. New plugin features, team tasks, and access through Slack and Microsoft Teams bring that work closer to the tools employees already use. OpenAI also announced privacy options, reviews, and approval controls as these systems take on more responsibility. Some products are available now, some are limited previews, and some are still coming. A company should check its plan and the feature's actual availability before building a process around it.

The tempting response is to connect everything and tell an agent to help the team. That is also where the hard questions begin. Consider a 30-person business whose operations lead spends every Monday assembling a status update. The information sits in a project board, customer messages, a spreadsheet, and the heads of two managers. An agent could collect changes and prepare a draft while the team works on other things. But it still needs to know which numbers are final, which customer issue is sensitive, what counts as an exception, and who can approve the update before it goes out. Those are business decisions, not settings a model can infer reliably from a product demo.

At Ruby, this is why we start by following the work with the employees who do it. We document the recurring questions, the places people look for answers, the handoffs that slow a process down, and the decisions that belong to a person. From there, a team can define a narrow job for an agent: gather the source material, flag missing information, produce a draft with links back to its evidence, and stop for review before anything is sent. A leader can then judge whether the workflow actually saves time, improves the quality of the update, or simply moves the same review burden somewhere else.

Training employees to use that workflow is part of the job, but the work cannot end with a launch session. Platforms change, permissions change, and a new feature may make a step easier or introduce a new decision for the team. Through its ongoing support packages, Ruby helps employees understand those updates, decide whether they fit the tools the company already uses or call for a different platform, and adjust the workflow with the people who own it. What worked at launch needs to keep working as the ecosystem changes.

The human part gets more important as the software becomes more independent. An always-on agent needs a clear owner, limited access, a way to inspect what it did, and a way for an employee to correct it. OpenAI says dots have permission and approval controls, and its DevDay recap describes new shared spaces and team tasks. Those features can make a responsible workflow possible. They do not decide the right boundary for a customer escalation, a finance approval, or an answer that could affect someone's job. The people closest to that work should help set those boundaries and test them in practice.