My journey of working with AI for productivity (as of May 2025)
I am a typical knowledge worker, working on business development, strategy, product discovery, product marketing, etc.
My usual deliverables:
Product presentations to the customers (PDFs)
Market research for our teams (Notion pages)
Strategy analysis to the leadership team (Miroboards + Notion pages)
Product discovery for the product and engineering (Notion pages)
I have juggled among different LLM applications: ChatGPT, Claude, DeepSeek, Gemini, Grok 3, before settling most of my work with ChatGPT.
My favorite workflow: starting with o3 deep research, doing several rounds of Q&A, and then finishing with a one-pager summary.
Why I settled with ChatGPT for now:
ChatGPT is the best product for my workflow.
It has a separate app on Mac, so I don't have to use the web app. This is a big deal because I have on average 12+ web tabs open.
Deep research is the best by far: more comprehensive and to the point.
o3's answers are relevant and straight to the point
The data connectors are not good at the moment. While every other LLM chatbox is not great either.
For a brief moment, Claude could have my business, but as a productivity product, it doesn't suit my needs - gathering and summarizing information.
Despite all the agent hype, for now, Deep Research is the only one that feels like a real one. I give it an objective, it delivers a satisfactory result most of the time.
There are still two main things that ChatGPT could improve: 1. work better with other data sources, 2. make the deliverable transfer easier. Transferring information from ChatGPT to GSlide or Canva is a drag.
The current AI evaluation benchmarks are disconnected from the working reality. Most knowledge workers don't solve hard physics equations. As a matter a fact, it functions a lot with feeling instead of bench numbers.
As of mid-2025, models are getting good enough, it's the product that needs improvement.