Agents in the current format are unlikely to go beyond a current levels of reliability. I believe agents are a good use case in a low trust environments (outside of coding where you could see the errors quickly with testing or deployment) like inter-company communications and tasks, where there are already systems in place for checks and things going wrong. Might be a hot space in some time. For intra company, high trust environment cannot just be a workflow automation given any error would need the knowledge worker to redo the whole thing to check if its correct. We can do it via other agents - less chances of it going wrong - but more chances it screws up in the same place as previous one.