I’m looking for prompt patterns or agent designs that are specifically built for complex, persistent information search on the web, with deep integration of human feedback along the way.
Here’s what I’m aiming for:
A prompt (or agent behavior) that treats research as a mission, not a one-shot task — it should pursue answers with focus, creativity, and adaptability.
When the AI encounters a limitation (e.g. login walls, unavailable languages, dead ends), it informs the human and suggests what help is needed (“log in here and search this”, “this forum likely has answers, but I can’t access it”).
It should explain its reasoning while it searches, including why it changes direction or chooses one path over another.
The agent should be willing to ask the human for clarification or confirmation, especially if alternative directions emerge.
It must persist with the task, not reduce output quality or give up due to difficulty or token usage.
Multilingual awareness is a bonus — many valuable sources are not in English.
I don’t want an auto-run bot like AutoGPT — I want an intelligent partner that actively collaborates with me on difficult research.
So my core question is:
Are there already proven prompts or prompt architectures like this? Maybe from real-world use, research, or toolkits like ReAct, Reflexion, Langchain agents, etc.?
If there are examples (successful or failed), I’d love to study them.
Thanks in advance! I’m also happy to share my own architecture draft if someone’s curious.
All “hands off” vibe-coding systems (Claude Code, Bolt, Lovable, etc) inevitably hit the 70/30 problem, which is solved with a more hands-on approach using tools like Cursor, along with the technical experience to understand and continue.
This paradigm can be used in other fields like video-making, and yes, research.
The fact that nobody has created a research agent as an assistant tool is astonishing. There’s been many times I just wished I could help point the model to the right direction, or say “hold on, this article is trash”. One of my biggest difficulties is trusting the model’s ability to stay in the time period that I’ve instructed (it never does, yet happily lies and says that it is)
The backend systems like “deep research” that exist all are single-shot systems from a “users” perspective (i.e. the process of the research happens independently of the user, and THEN the output is returned to user) [even though “under the hood”, it’s actually a multi-turn system for the LLM]
It doesn’t matter how good your prompt is, your working within the confines of what others have programmed/developed at the middleware level (the application itself, the LLM itself, etc.).
@aprendendo.next is correct, what your describing is actually a multi-shot recursive RAG system (likely with automated context window modification/truncation). This is absolutely non-trivial and highly complex.
It’s a beautiful idea (what you described) that I’m sure many others share, in fact it’s very likely a platform for such activity will be released in the upcoming months from someone.
Building an MVP for such a platform, especially open source, would be pretty awesome.
If your interested in doing something in terms of code/development to help make your idea reality, this is a good place to start.
Otherwise, you’ll have to keep waiting for someone to actually develop the tool your talking about, which is totally non-trivial and will require actually a relatively massive backend that also requires complex “guidelines” (pre-loaded or user input) sets of instructions/decision tree to allow the LLM to operate as a back-and-forth in the way you are describing.
Then a little explanation.
I realize the limitations of interacting with a certain kind of AI, but there are prompts that work brilliantly in these conditions because they are built with a deep understanding of AI
I received such a promt for information processing as a gift and thought that there are similar well-known solutions for complex web search.
That’s where my question comes from
So my core question is: