Ok @herdrdev is very cool. I wanted to try out the ability to control it using agents themselves. I had a bunch of articles I had been meaning to read so in a herdr workspace I had @claudeai instance be my agent orchestrator. It's job was to create an agent that would research each article, dump its findings in a local file and report back to the main agent when done.
The orchestrator created new panes in herdr and watched their progress. Once each was done it did it's own review off all the research files and created a master document I can now read. Sick.
Watching a lot of teams jump into AI by letting everyone build their own thing. Custom GPTs, apps, agents, cowork schedules or half-finished automations sitting in someoneβs personal account.
Give that 6 months. Then someone leaves, or a model gets deprecated, or the API changes, and suddenly nobody knows whatβs running where or who owns it.
Worth thinking about the shared layer before youβve got 40 versions of the same workflow scattered across the team.
Ok this worked pretty well. Claude sessions was super easy to use and work with.
Next step is build a UI to see what the agents are up to and an easier place to edit prompts or settings from.
Working on a Cloudflare + Claude Managed Agents app that helps us spin up new agents using a shared foundation. It has skills to create new slack apps + bots, new claude agents, new tools, new skills and new cloudflare infra like workers or databases.
Hopefully streamlines future internal agent creation!
When can we expect some decent usage data for teams using claude code across CLI/Web/Desktop @ClaudeDevs? Analytics area is pretty pointless currently :/
Wouldnβt it make sense if Claude Managed Agents became a similar interface to ChatGPTβs agents? This would allow users to access their created agents directly within Claude rather than needing to invoke them via APIs.
Surely weeks away?