This post is great - https://aarontay.substack.com/p/how-agentic-are-academic-deep-research
I learnt a lot.
My main reflection is that the fixed workflows that the current crop of tools are showing were developed when models were weaker, yet from a product perspective, trying to build something that is simultaneously specific and general, is really hard.
Models are going to continue to improve in 2026. There was a moment towards the end of 2024 when we were unsure about that.
This tension between having immense power, and making it easy to use, is going to be an ongoing product tension over the next few years (not a novel thought). My thoughts and prayers are with all the researchers and product folk out there.
I learnt a lot.
My main reflection is that the fixed workflows that the current crop of tools are showing were developed when models were weaker, yet from a product perspective, trying to build something that is simultaneously specific and general, is really hard.
Models are going to continue to improve in 2026. There was a moment towards the end of 2024 when we were unsure about that.
This tension between having immense power, and making it easy to use, is going to be an ongoing product tension over the next few years (not a novel thought). My thoughts and prayers are with all the researchers and product folk out there.