AI-agent workflows often assume one user, one session, one agent.
But building with agents can involve multiple people: someone starts an experiment, another adds context, someone else redirects the agent, and another teammate takes over the task.
Having to restart the conversation each time can mean losing valuable context.
I’ve been exploring Sharkly.ai because it approaches AI-agent work as a collaborative workflow, where multiple people can contribute to the same session and hand work off when needed.
For teams experimenting with agents, research workflows, or AI development:
Would shared live sessions make working with AI agents more useful?
Definitely shared sessions could make agent workflows much easier for teams, especially when context needs to be passed between researchers and developers.
Shared sessions are already a real direction, so the question is less “would it help” than “which implementation”. Persistence and handoff are solved at the framework level (LangGraph checkpoints with a thread ID, OpenAI Agents SDK handoffs, Microsoft Agent Framework). On the product side, Claude Tag (shared agent in a Slack channel) and Zed’s multiplayer agents cover live shared sessions. The hard parts are elsewhere: who is allowed to redirect the agent mid-run, how concurrent instructions get resolved, and keeping an audit trail of who steered what.
Also, this is the same product (Sharkly.ai) as in several of your other threads here, and the reply right under this post is generic enough to look like part of the same push. Saying upfront that it’s your product, with concrete details on what it does that those tools don’t, would get you better answers than open-ended questions.
Shared AI-agent sessions could be particularly useful when tasks involve multiple handoffs, since preserving context can reduce duplicated work and keep everyone aligned. I think the most interesting part is how permissions, concurrent instructions, and session history would be handled without making the workflow confusing. It would be interesting to see how this approach develops for real-world collaborative AI workflows.