Agents Turn Simple Keyword Search Into Compelling Search Experiences

2 months ago

4 replies

These people really mis-understand how people like me use search. I don't want "an experience," I want a list of documents that match the search specifier.

softwaredougAuthor

2 months ago

3 replies

Author here, well specifically I'm using "experiences" to mean basically ranking / finding relevant results.

soco

2 months ago

2 replies

Wouldn't the "dumb API search" be already enough for users? I'm not sure my decades experience as Google search user has improved once they leaned heavy on the ranking and adding "relevant" things...

2 months ago

> once they leaned heavy on the ranking

Not sure what you mean here. Google started with PageRank which is a decent, albeit gamable, ranking algorithm. They’ve never not been leaning heavy on ranking.

Mid 2010s, Google began supporting conversational (non keyword) searches. This is because a good number of queries are difficult to reduce to keywords. But it is an inherently harder problem. And at the same time, the open web enahittified itself, filling up with SEOed blogspam. And a lot of user generated content moved into walled gardens.

about 2 months ago

> Wouldn't the "dumb API search" be already enough for users?

This has been deprecated a long time ago (10 years ?). It hurts (ad) sells.

about 2 months ago

You mean hiding relevant results. Not only that today's search engine gives you mostly crap, they also insert "results" which have no relation whatsoever with your search term.

nottorp

2 months ago

Well, at least you didn't use 'engagement'.

Edit: oops, you used it in another comment on here.

awlejrlakwejr

2 months ago

1 reply

Absolutely. We need separate interfaces for work and play.

2 months ago

You might be onto something here. What we've lost I notice mostly when working. When I search for "Bizarre undocumented compile error B7102343. Rebooting Universe." I really want to see ONLY those pages that have these keywords. But when I'm trying to find a list of DMV offices near me, I want the "search" engine to interpret my question "where are DMV offices near me?"

Maybe it is two different products: index search and fuzzy search?

2 months ago

1 reply

I agree, I agree, I search in Google, but Gemini is quite good and most of the times the answer is correct or close enough to save a lot of my time.

inetknght

2 months ago

1 reply

> I search in Google

What I understand you to mean is: you ask Google to give you the most-advertised product and hope that it's what's best for you.

2 months ago

1 reply

I skip the "sponsored" links at the top. For programming questions it used to link to the most relevant Stack Overflow question and the exact doc of that language (that many times are difficult to navigate). It autocorrected typos and used synonyms like "pause" -> "sleep".

krageon

2 months ago

It isn't like the sponsored links are the only paid-for high listing in your results, SEO is a thing that exists and has nothing to do with quality. Just money and time.

about 2 months ago

Most of those people substituted their intelligence with AI.

dsr_

2 months ago

1 reply

"Agents, however, come with the ability to reason."

Citations needed. Calling recursive bullshit reason does not make it so.

softwaredougAuthor

2 months ago

1 reply

Well call it what you want, I'm referring to the reasoning functionality of GPT-5, Claude, etc. It seems to work pretty well at this task. These tools prefer grep to some fancy search thingy.

2 months ago

1 reply

> These tools prefer grep to some fancy search thingy.

Passing on the cost to you. The fancy search thingy could navigate the AST and catch typos.

softwaredougAuthor

2 months ago

1 reply

It's more accurate to say the agents like simple, interpretable tools. Clear input / obvious output. That might be more an AST thing, keyword search, search with very straightforward filters, etc.

I don't thing they do well with search that is built for human engagement, which is a more complex tool to reason about.

2 months ago

1 reply

It makes no different to the agent. The response from grep is a list of matches, and so it is from something more intelligent. A list is a list.

> I don't thing they do well with search that is built for human engagement, which is a more complex tool to reason about

I agree! Structured outputs are best.

shadowgovt

2 months ago

Under the hood, LLMs are vector analysis engines. They benefit from spaces where the result of change is smooth.

Adding levels of indirection and secondary reasoning to the search interface makes the results less smooth. This is one of the things humans complain about often when using e.g. Google: "I'm interesting in searching for X, but all these results are Y." Yes, because X and Y are synonyms or close topics and Google is mixing in a popularity signal to deduce that, for example, your search for `tailored swift database` is probably about a corpus of Taylor Swift song lyrics and not about companies that build bespoke Swift APIs on top of data stored in Postgres.

If you're driving the process with an LLM, it's more of a problem for the LLM if it's searching the space and the search engine under it keeps tripping over the difference between "swift means a programming language" and "swift means a successful musician" as it explores the result space. A dumber API that doesn't try to guess and just returns both datasets blended together fits the space-search problem better.

the_snooze

2 months ago

3 replies

If agents are making value judgments ostensibly on my behalf, why should I trust them to continue to be aligned with my values if in practice they're almost always running on someone else's hardware and being maintained on someone else's budget?

We'd be stupid to ignore the last 15+ years of big tech "democratization"-to-enshittification bait-and-switch.

2 months ago

That was never not the case. There are always value judgements. Even for keyword searches, there are often hundreds of exact matches, and one you want might not even be an exact match.

This article is about how Target can use LLMs to help you find patio furniture. I guess you could imagine the LLM upselling you on a more expensive set?

evilduck

2 months ago

Unless you're running your own search engine for yourself, search indexes and vector databases already manage what data they want to ingest, they contain rank weights, keyword aliases, and prefiltering for controlling the search result in favor of the service provider's desired outcome. And these all run on someone else's hardware maintained on someone else's budget.

Adding an LLM or agentic data flow and a tuned prompt to the mix does nothing to change who is in charge, it was never you.

softwaredougAuthor

2 months ago

Search engines currently do this for better or worse. But they still want you to buy products.

The bigger issue is I’m not sure agents are trained to understand what users find engaging. What makes users click.

jillesvangurp

2 months ago

1 reply

Interesting approach. It might be helpful to give the agent more tools though. Some simple aggregations might give it a notion of what's there to query for in the catalog. And a combination of overly broad queries and aggregations (terms, significant terms, etc.) might help it zoom in on interesting results. And of course, relatively large responses are not necessarily as much of a problem for LLMs as they would be for humans.

softwaredougAuthor

2 months ago

I agree, my post only scratches the surface. I want to give more knobs to the agent. But not so many that it can't really experiment / reason about them.

intalentive

2 months ago

1 reply

The doc string becoming part of the prompt is a nice touch.

It seems plausible and intuitive that simple tools dynamically called by agents would yield better results than complex search pipelines. But do you have any hard data backing this up?

softwaredougAuthor

2 months ago

I don't have any hard data, no. That will be the other-other side project I may or may not have time for :)

2 months ago

1 reply

Prompts for this:

Turn this into a paragraph-sized prompt

Turn this into a n document length formal proposal,

And then split that into paragraph sized token optimized prompts

View full discussion on Hacker News

2 months ago

Aren't "Attention is all you need" Transformers good at suggesting suggested search terms?

rgavuliak

2 months ago

Why do you expect an LLM to provide an accurate distance metrics?

pimlottc

2 months ago

“thick-daddy search API”…

hakfoo

2 months ago

Keyword search is the compelling experience.

I want my machine to be determinstic and non-magical. I am so tired of search tools that won't let me actually search for what I want because it clearly thinks I meant something else.

daft_pink

2 months ago

I think Comet's search is really nice and worth the $20 a month, but not $200 a month that it currently costs although it is a little slow. My experience is similar to this article.

mips_avatar

2 months ago

One interesting thing about the lack of click stream feedback is that you can generate it synthetically. If you've got your model evaluating the quality of search responses and adjusting its queries when there's a miss, you get to capture that adjustment and tune your search engine. In user click search you need scale to tune search, now you can generate it. The only problem is you need to trust your agent is doing the right thing as it keeps searching harder.

ID: 45347363Type: storyLast synced: 11/20/2025, 12:29:33 PM

Want the full context?

Jump to the original sources

Read the primary article or dive into the live Hacker News thread when you're ready.

Read Article View on HN

Last activity about 2 months agoPosted Sep 23, 2025 at 10:17 AM EDT

Agents Turn Simple Keyword Search Into Compelling Search Experiences

softwaredoug

67 points

35 comments

Mood

skeptical

Sentiment

mixed

Discussion Activity

Very active discussion

First comment

20m

Peak period

Day 1

Avg / period

11.7

Comment distribution35 data points

Loading chart...

Based on 35 loaded comments

Key moments

01Story posted
Sep 23, 2025 at 10:17 AM EDT
2 months ago
Step 01
02First comment
Sep 23, 2025 at 10:37 AM EDT
20m after posting
Step 02
03Peak activity
29 comments in Day 1
Hottest window of the conversation
Step 03
04Latest activity
Sep 29, 2025 at 2:59 AM EDT
about 2 months ago
Step 04

Generating AI Summary...

Analyzing up to 500 comments to identify key contributors and discussion patterns

Discussion (35 comments)

Showing 35 comments

2 months ago

4 replies

These people really mis-understand how people like me use search. I don't want "an experience," I want a list of documents that match the search specifier.

softwaredougAuthor

2 months ago

3 replies

Author here, well specifically I'm using "experiences" to mean basically ranking / finding relevant results.

soco

2 months ago

2 replies

Wouldn't the "dumb API search" be already enough for users? I'm not sure my decades experience as Google search user has improved once they leaned heavy on the ranking and adding "relevant" things...

2 months ago

> once they leaned heavy on the ranking

Not sure what you mean here. Google started with PageRank which is a decent, albeit gamable, ranking algorithm. They’ve never not been leaning heavy on ranking.

about 2 months ago

> Wouldn't the "dumb API search" be already enough for users?

This has been deprecated a long time ago (10 years ?). It hurts (ad) sells.

about 2 months ago

You mean hiding relevant results. Not only that today's search engine gives you mostly crap, they also insert "results" which have no relation whatsoever with your search term.

nottorp

2 months ago

Well, at least you didn't use 'engagement'.

Edit: oops, you used it in another comment on here.

awlejrlakwejr

2 months ago

1 reply

Absolutely. We need separate interfaces for work and play.

2 months ago

Maybe it is two different products: index search and fuzzy search?

2 months ago

1 reply

I agree, I agree, I search in Google, but Gemini is quite good and most of the times the answer is correct or close enough to save a lot of my time.

inetknght

2 months ago

1 reply

> I search in Google

What I understand you to mean is: you ask Google to give you the most-advertised product and hope that it's what's best for you.

2 months ago

1 reply

krageon

2 months ago

It isn't like the sponsored links are the only paid-for high listing in your results, SEO is a thing that exists and has nothing to do with quality. Just money and time.

about 2 months ago

Most of those people substituted their intelligence with AI.

dsr_

2 months ago

1 reply

"Agents, however, come with the ability to reason."

Citations needed. Calling recursive bullshit reason does not make it so.

softwaredougAuthor

2 months ago

1 reply

Well call it what you want, I'm referring to the reasoning functionality of GPT-5, Claude, etc. It seems to work pretty well at this task. These tools prefer grep to some fancy search thingy.

2 months ago

1 reply

> These tools prefer grep to some fancy search thingy.

Passing on the cost to you. The fancy search thingy could navigate the AST and catch typos.

softwaredougAuthor

2 months ago

1 reply

It's more accurate to say the agents like simple, interpretable tools. Clear input / obvious output. That might be more an AST thing, keyword search, search with very straightforward filters, etc.

I don't thing they do well with search that is built for human engagement, which is a more complex tool to reason about.

2 months ago

1 reply

It makes no different to the agent. The response from grep is a list of matches, and so it is from something more intelligent. A list is a list.

> I don't thing they do well with search that is built for human engagement, which is a more complex tool to reason about

I agree! Structured outputs are best.

shadowgovt

2 months ago

Under the hood, LLMs are vector analysis engines. They benefit from spaces where the result of change is smooth.

the_snooze

2 months ago

3 replies

We'd be stupid to ignore the last 15+ years of big tech "democratization"-to-enshittification bait-and-switch.

2 months ago

That was never not the case. There are always value judgements. Even for keyword searches, there are often hundreds of exact matches, and one you want might not even be an exact match.

This article is about how Target can use LLMs to help you find patio furniture. I guess you could imagine the LLM upselling you on a more expensive set?

evilduck

2 months ago

Adding an LLM or agentic data flow and a tuned prompt to the mix does nothing to change who is in charge, it was never you.

softwaredougAuthor

2 months ago

Search engines currently do this for better or worse. But they still want you to buy products.

The bigger issue is I’m not sure agents are trained to understand what users find engaging. What makes users click.

jillesvangurp

2 months ago

1 reply

softwaredougAuthor

2 months ago

I agree, my post only scratches the surface. I want to give more knobs to the agent. But not so many that it can't really experiment / reason about them.

intalentive

2 months ago

1 reply

The doc string becoming part of the prompt is a nice touch.

It seems plausible and intuitive that simple tools dynamically called by agents would yield better results than complex search pipelines. But do you have any hard data backing this up?

softwaredougAuthor

2 months ago

I don't have any hard data, no. That will be the other-other side project I may or may not have time for :)

2 months ago

1 reply

Prompts for this:

Turn this into a paragraph-sized prompt

Turn this into a n document length formal proposal,

And then split that into paragraph sized token optimized prompts