V2 was the major fix to the shortcomings of V1.
Templates were introduced, and separated from the agents themselves. Also aside from templates, checklists were introduced to give more power in actually vetting the the documents or artifacts being produced were valid and of high quality through a forced round of advanced elicitation.
During V2, this is where the discovery of the power of Gemini Gems and Custom GPTs came to light, really indicating how powerful and cost effective it can be to utilize the Web for a lot of the initial planning, but doing it in a structured repeatable way!
The Web Agents were all granular and clearly defined - a much simpler system, but also somewhat of a pain to create each agent separately in the web while also having to manually export and reimport each document when going agent to agent.
Also one confusing aspect was that there were duplicates of temples and checklists for the web versions and the ide versions.
But - overall, this was a very low bar to entry to pick up and start using it - The agent personas were all still pretty self contained, aside from calling out to separate template files for the documents.
