Great questions here, Mike. I’ll do my best to provide some insight from what I’ve experienced:

1) I think where we’re headed is AI-agent-focused platforms that allow you to create custom agents or use ready-made ones. Mastering how to design and guide these agents, often by writing system-level prompts, is going to be the key skill as these tools evolve and become central to workflows. Another thing I would recommend is to try out frameworks like Agent Zero (github.com/frdel/agent-…). This will allow you to really get in there and see how an agent works and how to guide it though APIs, prompting, etc. This is where I started. I literally had no idea how to deal with an agent, but anyone can learn. Even with all of this talk with OpenAI’s Operator, there are so many other developments and frameworks taking place with less attention but will probably be more valuable.

2) I have written a lot of big and complex prompts for LLMs. Some covered on AI Disruptor, mostly for content creation. I also worked on the inside of some startups developing AI agent platforms, and what some people don’t realize is a lot of these same prompts are very similar to the system instructions for an agent. Think of prompts that ‘program’ personality traits, skillsets, etc. I can’t speak for how some of the other agents are started, but with time we will learn a lot more into the process. I think 2025 will see a lot of this covered in my newsletter. We don’t actually have that many agents running around out there yet to know.

3) That is a good question that I don’t know the answer to yet. I think a main part will just be educating people on analyzing the output and spotting hallucinations, etc.

4) I think by the end of 2025. And like I mentioned above, we don’t even need to focus on Operator or computer use because there is a lot more groundbreaking stuff happening outside the big LLM companies IMO.

I hope some of this helps.