Hi @Davidkm 
Wanted to follow up with something that may be relevant to your company-wide rollout. Retool just opened early access to a centralized agent harness:
It’s aimed at the “one agent for the whole company” scenario you described. Employees use agents directly in Slack, Teams, or Google Chat, while admins control which tools and data the agents can reach. Spend and activity are traceable by person, model, and job, which could help when you’re deciding how to scale usage across many people.
On the memory side, here’s an example of how Retool uses this internally: our shared data agent, Dax, saves corrections from teammates to its memory so future answers improve. It’s a similar idea to the context problem you raised.
Since you’re deciding where to invest your time, joining the waitlist could be a low-effort way to see whether it fits. 