Currently a lot of MMGIS' new documentation is tailored for an Agentic Developer, one setting up MMGIS, developing new plugins, and so on. Another persona an agent may take is that of an administrator and the agent-specific documentation for that role is far less developed. For instance: asking AI to stand up a new mission and just include the following DEM (https://zenodo.org/records/22164484) is not as smooth as it could be:
- The agent does not know it has to look at a series of plugin jsons to know how to construct a config json object.
- The agent does not immediately know that it should be in a non-coding capacity
- While the agent knows how to setup MMGIS, over longer sessions it often forgets its admin password to the /configure and has to waste time trying to reset it in the database
- How to configure projections was less clear to the agent in the case of the DEM above (it did not know Venus projection tilematrixset were already provided via planetcantile)
- and so on.
I good way to generate adequate documentation for an administrator roles may be to have a lead agent spawn child agents in waves and give each child agent the task of setting up a different new mission with different dataset and report back pain points and keep iterating until child agents can, near-zero shot, quickly configure a variety of new and existing missions.
Currently a lot of MMGIS' new documentation is tailored for an Agentic Developer, one setting up MMGIS, developing new plugins, and so on. Another persona an agent may take is that of an administrator and the agent-specific documentation for that role is far less developed. For instance: asking AI to stand up a new mission and just include the following DEM (https://zenodo.org/records/22164484) is not as smooth as it could be:
I good way to generate adequate documentation for an administrator roles may be to have a lead agent spawn child agents in waves and give each child agent the task of setting up a different new mission with different dataset and report back pain points and keep iterating until child agents can, near-zero shot, quickly configure a variety of new and existing missions.