The Gartner Hype cycle is a well known technology trope:
However, I like to change it and call it the AI Try Cycle:
To keep on top of all of the AI tools out there would be really, really hard to do unless it was your full time job.
So each cycle, I’ll pick a few tools and work on getting really good at them.
Here’s my last cycle breakdown:
Writing Code
Cursor.com : It took my previous coding environment, VS Code, and added integrated AI coding (with OpenAI or Claude etc engines). For many one-off programs that I do for work, worth the $20 a month - so much so that I cancelled Claude.
Replit: Making websites in a snap AND hosting too, which is a problem I have with lovable even though I like lovable’s end result and coding philosophy (pushing to Github instead of ‘magic’) better).
Example of Replit:
Example: https://taskanalyzer.replit.app/ lets you enter tasks (right now without a login for the sake of this blog post) and automatically, through an OpenAI API connection, sorts by ability to be increased in efficiency with AI and on the next page, even is able to talk about how to get started trying things.
Next up:
Browser-use Agents. I’m paying for ChatGPT PRO ($200!) for a month to see how well Operator it works as well as o1 pro. In the meantime I’ll see what I can do to get Browser-Use agent working really well either with openai-GPT-4o or perhaps deepseek or qwen that I’ve seen other folks doing.
Reasoning Chat Models: Again will be using o1 and/or o3 as they are released, but also have installed Deepseek locally running on my ollama+Open WebUI instance. Creepy how it talks to itself but maybe it will sound less like Gollum in future iterations.
Open WebUI: I’ve been using this tool for a half dozen months, but need to really dig into creating ‘pipelines’ and see if I can get the PhiAgent library working well so I can run agent chats inside of it instead of Streamlit etc.. (Recent post on how to install it)
