AI assisted search-based research actually works now

most of what LLMs are being sold for, generally only work for cherry picked examples.

but there is a growing set of tasks that (beyond meme generation) that are proving to be helpful tools as part of a larger task.

deep search / online search seem to fit the bill.

in some sense, this is automating the work done by folks who wield a search engine like a scalpel.

i think this essay opines too optimistically about the results (looking up documentation works for me, but not generating code based on that documentation). and those results likely don't map to the average experience.

with that said, i do see LLMs improving at their ability to summarize in a focused way and pull together resources in a way that it helps me focus on what to read/research next. but if you push the tool to do the full task for you, you are going to be sorely disappointed unless your task is truly simple.

overall, exciting times.

In the first half of 2025, AI systems have crossed the line into being genuinely useful for search-based research tasks. OpenAI's o3 and o4-mini, as well as Google Gemini 2.5 Pro, are highlighted for their ability to provide useful answers grounded in search results without hallucinations. The author shares examples of using these tools for tasks like porting code to a new library version. The rise of AI-assisted search raises questions about the economic model of the web, as users may increasingly rely on chatbots instead of visiting websites directly.