Hacker Newsnew | past | comments | ask | show | jobs | submit | freediver's commentslogin

this kind of randomness is why ai is never going to beat humans


Because Kagi plans are/were priced for human use, and programatic api use/agents can easily 100x that, performing dozens of searches for one user query. So either Kagi plan price increases and API calls then get included in the price (but then this impacts all users even those not using agents) or there is a pay what you use mechanism for API (which there is).

Agentic AI users are familiar with usage caps and rate limiting.

Any API use given away for free eats into Kagi's (already slim) margins, and unlike VC funded businesses, for Kagi money has to come from users or Kagi doesnt exist. It is a reality of running a user-funded business. If the user doesn't pay for something, Kagi has to.

I think the point is that it would be nice if people who are already paying for human-driven searches could also get an API key included in that plan.

Right now, API usage is only available using a PAYG credit system, even if you've already got a subscription. But giving subscription users an allotment of free credits that covers light API usage would be nice - I could plug it into my agent of choice for casual chats or exploration, while still having to buy credits for more extreme or production usage.


It would be nice indeed, but it's a bit like wanting to share a gym subscription when you're not using it. The economics would fall apart if everyone used 100% of their subscription, unless sharing was priced in from the beginning.

The city transport system where I live has a subscription pass you can buy that is normally tied to a single person. However, if you pay a little bit extra you make the ticket transferable, and even let someone else travel with you for free on weekends and in the evenings. Like you say, they've presumably priced in the extra usage there that comes from it being transferable, or from the extra travellers, but also put limits on how much leeway you have to abuse the system.

I'm imagining rate limits that would prevent the user from making many more API calls than they would have made searches anyway. If you want unlimited 24/7 searches, you can pay for the credits directly, but if you just want to partially replace your existing search habits with e.g. interactive agent sessions, then that would be fine.

A more specific example would be how Anthropic allow certain Claude subscriptions to get free API credits, with some restrictions on the age of the account and with the number of credits being limited.

I think there is a workable solution here, although I can imagine that figuring out the details (how many credits to offer, what sort of limits, how to price the whole thing, etc) is complicated enough that Kagi have decided it's not worth the effort for the small number of users for whom this feature would be useful. And generally I'm very satisfied with Kagi, so I'm not complaining here or trying to suggest that this is something that they should be expected to offer. It would just be nice!


What if the usage were shared between the two and a single API search used more of your quota than a UI search? There's no reason the economics can't line up.

e.g. if you expect the average user to use 50% of the search quota / month through the UI, but 100% through the API, then you make the quota shared between the two but double the effect of usage from the API.


That would work if the average user with Search + API used the same "amount" (Costs to Kagi) as the average account using Search only. If it nets out to a higher cost, Kagi is making less out of the same subscription amount.

Especially if the API induces demand.

(Don't shoot the messenger, just trying to explain why it makes sense that they're not doing this)


I think the only way to make it not harmful to their finances would be if they lowered the cap (even for human use) to where heavy users and API users alike ended up paying 'overages' constantly. Everybody who has studied consumer behavior generally finds that people hate overage charges to an irrational extent[1] so that would be a terrible move compared to just saying "API users, you've got to pay extra, sorry."

[1] Meaning, they are more satisfied with an 'unlimited' or just way-more-than-you'll-ever-need option, even when it costs them more when averaged over even a few months. I'm not saying this makes you an irrational person -- I have the same illogical feelings myself.


That's not really illogical, it's just because many companies use the fine print to fleece customers that are slightly above the limit. Think Ryanair. Or mobile roaming.

This is why unlimited or at least a service stop instead of overage is highly preferable. It's the reason I only use prepaid mobile plans too so they can never overcharge me (and it's cheaper than a contract too)


Is the same not true of 50/100 free browser-based searches?

No, and it's a real problem for free and trial tiers for services like this.

When a real user is going through the trial period, those 50 searches get them familiar with the tool. When their usage runs out, they're forced to decide if they want to pay to continue or invest the work into replacing it.

AI agents typically don't care. They may sign up for a trial, use the trial, and then when the trial runs out they might get creative and create a new account, or silently switch to another service to avoid bothering the user.

Browsing some of the forums where people discuss their search setups is enlightening. The current favorite strategy is to have your AI set up access to all of the different search providers with free tiers and then rotate among them to avoid having to pay for anything.

Defeats the purpose of the free tier.


> Browsing some of the forums where people discuss their search setups is enlightening.

Interesting, is this for personal use or is it some business application? In any event I'd imagine the same can be done via agents using a browser.

> AI agents typically don't care. They may sign up for a trial, use the trial, and then when the trial runs out they might get creative and create a new account, or silently switch to another service to avoid bothering the user.

That is a crazy amount of compute just to conduct a search.


If burning tokens is free then the user doesn’t care. None of this changes until free money stops flowing into AI companies.

The people on AI company plans don't need these services because they're included in the plans.

It's the people downloading open weight models and running OpenClaw (or this month's trend) who don't care how many tokens it takes because it only costs them electricity.


> That is a crazy amount of compute just to conduct a search.

Real businesses aren't doing stuff like this.

It's the home users who use LLMs as a hobby. Some power users pay for Exa and other services. A lot of home LLM users will go to unbelievable lengths to not pay for anything that isn't hardware. Seeing what your agent can do is half the fun for them.

Set one of them against a goal and tell it to not bother you until it has the answer. They will do some weird things as they go.


Only in the trivial sense that they also cost.

But those are an offer and are used to advertise and get more users. Not to mention, as the "50/100" says, they're capped.


> Only in the trivial sense that they also cost.

I don't get it, what other sense is there? Seems like 50 daily API queries would be lighter weight than 50 browser queries. And with an API you can require a key (aka subscriber).

> But those are an offer and are used to advertise and get more users.

"Try our search engine from your browser" / "Try our search engine from your agent setup" These both seem like user acquisition pitches. The real difference with programmatic access is that you'll likely get users hitting the cap and paying for more usage if they like the product.

> as the "50/100" says, they're capped.

... which is exactly what I was suggesting for API use.


Why would API searches be lighter? The searches are expensive, not the rendering.

I guess because a web-based search is fundamentally one or more API calls so everything before and after it is expense. Maybe it's very near zero.

>I don't get it, what other sense is there?

Obvioulsy the TYPE and RETURNS of some cost.

>These both seem like user acquisition pitches. The real difference with programmatic access is that you'll likely get users hitting the cap and paying for more usage if they like the product.

50 searches from the web feel way more generous to a user than using the agent (most wont even hit them), hitting the 50 cap in like 10 minutes, and being negative about the whole experience.

>... which is exactly what I was suggesting for API use.

Which is exactly why it wouldn't work well for API use with such cap.


This comes out of marketing budget for new user acquisition.

Insurmountable then.

Or not a problem to begin with :)

>Any API use given away for free eats into Kagi's (already slim) margins

I refuse to believe there is a negligible margin decrease for agents using the regular search budget. Searching is cheap.


It’s about repackaging. With free/bundled API pricing, resellers could provide agents that min/maxed Kago API tokens in the background, eating Kagi’s compute abd effectively repackaging a competitor that deterred new human subscriptions.

That’s a bad business decision in the same way it would be a bad business decision for Netflix to re-enable infinite account sharing.


I'm not sure what you are trying to say. We are talking about a tiny amount of compute.

lol wut

They already include capped usage on assistant and translate, doesn’t seem like a huge ask to do the same for API.

Because Kagi plans are/were priced for human use

A bit off-topic, but when I signed up for Kagi, I naturally became more aware of how many times I use search on the internet, since I didn't want to run into the limit for my tier.

I assumed I was a heavy search user, but now that I watch myself, I'm really surprised how little I actually search anymore. I thought it might be a hundred times a day. In reality, it's closer to four or five times a day.

I don't know if it's my changing habits; or the fact that much of the information on the internet has become siloed, so I know where to go without wandering around; or some other factor.

I was just really surprised.


I think most of us might be surprised too. Searching 'The Whole Internet' used to be so useful. Now, I know most of the names of the sites that might hold the answer to my question, since many sites are either gone or pivoted to intrusive ads and paywalls. So now, often I could just go to the right site directly. e.g. genius dot com, some specific vendor's website, justwatch dot com, etc. And on top of that, I know what questions AI will nail much better ("What's the episode of Friends where Joey is scared by a spider?"). And adding to all of that, the sheer degree of degradation the search engines' indexes have suffered. Google can tell you the right website for, say, the city government of Philadelphia. So it has some use. But if you search for most open-ended topics, you're just going to get 1000 AI-written content farm sites with nonsense domains.

Genuinely wonder what would happen if all DDG users did this?

The extension is open source and macOS changes do interfere with it working. Apple is not making it easy. It is one of the reasons we created Orion - same lightweight, native Safari experience but integrated with Kagi.

https://orionbrowser.com


Orion is really good, especially for people who appreciate Safari.

Small web results for 'hacker news' right now on Kagi Small Web:

https://kagi.com/smallweb/?search=hacker+news


> For example, he's currently recording a bunch of heavy metal albums.

Oh, any links?


Based on this video I would believe you if you told me he was 69. This is unbelievable for 95. https://www.youtube.com/watch?v=TaButvSxPWQ

I've got a copy of his first album, The Transformed Man, which is quite entertaining to listen to. I have heard him do a cover of Black Sabbath's "Iron Man".

"Seeking Major Tom" might be of interest to you.

Google sez:

William Shatner’s heavy metal album, titled What the F Is Heavy Metal?, does not have an exact release date yet, though it is tentatively scheduled for winter 2026 or 2027 via Cleopatra Records.

Album Details

• Title: What the F Is Heavy Metal? • Band Name: The *uckers • Expected Release: Winter 2026 / 2027 • Label: Cleopatra Records • Format & Content: Features a mix of original tracks and heavy metal/hard rock covers (including Black Sabbath, Iron Maiden, Judas Priest, Metallica, and Ozzy Osbourne). • Notable Collaborators: Rob Halford, Tony Iommi, Zakk Wylde, Brian May, and Dave Lombardo, among dozens of other rock and metal musicians. • Live Debut: Shatner and his band debuted songs from the project live at Riot Fest in Chicago in September 2026. • William Shatner on Instagram Ladies, gentlemen, SWs fans… I have explored space. I have explored time. Now… I explore distortion. Yes. You read that correctl... • USA Today William Shatner debuts heavy metal band at age 95 At 95, William Shatner is starting a heavy metal band Greta Cross USA TODAY Sept. 17, 2026Updated Sept. 18, 2026, 9:14 a.m. ET Wil...


Not Shatner, but I need to drop that Patrick did a backing track on Brent Spiner's album. It has to be in here somewhere, and this seems as good a place as any.

https://www.youtube.com/watch?v=pFwiLa3TYwg&list=PL1T3N4t2kU...

> Sunspots: LeVar Burton, Michael Dorn, Jonathan Frakes, Patrick Stewart.

uNF!


Pretty remarkable (in the age of AI) that search/ad revenue actually grew 12%. Anyone have a rational explanation why is that the case?

For me Google has stayed as the default interface, if I need to ask a basic query (probably searches that don't yield much revenue but keeps me on Google), the results were becoming worse a 1-3 years ago but now the AI parts of Google is now good-enough(TM) that I mostly don't need to go to a separate interface, as long as users aren't fleeing they can probably continue to get paid for ADs (also the if YT ads are included, those have really become obnoxious but probably well monetized).

But you are not clicking on ads? And they aren't (yet?) in AI overviews?

Not more than accidentally or when actually interested in something that pops up, but as a friend said, I'm not a usual core customer so I've never been the baseline.

The thing they're doing correctly right now imho, is that AI mode is mostly toggled for things that passes a quick questions filter (probably low probability of a purchase, like "global illumination methods" giving a decent overview of the field) while if you just type green shoes it'll put out an assortment of online and local shopping options.

Now if as someone said in the sibling comments, Google is gouging their customers with the new auction model that's another problem, but if I was a shareholder I'd be less worried about an exodus of end users to OpenAI,etc than I was a year ago.


I saw a post on X saying Google changed the ad auction pricing model, but I don’t know much about this to comment if it’s accurate.


That post is shocking and probably deserves its own HackerNews discussion.

Thanks, this does make sense.

> that search/ad revenue actually grew 12%

it grew compared to q3 2025, but it is flat (0% grows) compared to q2 2026.


The money is mostly paid by companies wanting to find customers and the AI companies like OpenAi and Antropic don't provide much of that. The biggest competitor is probably Facebook who I'm not sure are really booming at the moment?

Curious what was the source of faith you previously had?

I was promoting Google models (more like Google ecosystem) for what Google was doing with their models. Making it accessible to non-technical people (which is what I am doing).

e.g. NotebookLLM was one that got me interested in Google's ecosystem. It was easy to use and explain even to my parents.

Now, their models (and NoteBookLLM merging to Gemini, making it extremely hard for non-tech people and other stuff) release very slowly, their models lie their ass off, very politically skewed, etc.


Being paywalled is not an issue (better than ad-ridden), but not being able to have micro transaction in a browser with one click still is mind boggling. As a creator of a browser, taking some blame for that.

I can do that. I can click on my credit card autofill and then "pay", so two clicks.

But sites won't let me do that, they want me to make an account first and they want to verify my email address and phone number which means leaving my browser and restarting the process each time I do because the browser got unloaded and the website doesn't save state. Almost all of the friction here is the fault of the website, not the payment process.

Usually I just can't make an account on a mobile device because of the browser unloading, so if it's important I do it later with my PC and if it's not important I give up entirely.


Well, a micro-transaction on your credit card isn't a great solution, since the base fee is something like $0.10 per transaction. You'll want a system to fill a low-transaction-fee wallet and then distribute from there.

Youtube Premium is a pretty good system. I pay $16/mo, and 55% of that gets distributed to creators based on my watch time.


There are lots of ways to do the tech without a $0.10 fee. I think the issue is more the websites don't ask for that, probably because $10 a month in one easy annualised payment of $120 pays better.

I'm not sure if there's an actual website that did it? I've made micropayments for other things like PAYG phones and speculating on crypto.


There were attempts to bring micro-transactions to the browser via W3C standards - but guess who obstructed it? Yes, the big G that's running the browser with most market-share (Chrome) and who's entire business-model it having your data so only they can be the middle man between you and ads. Imagine if people could just support what they like with one click, instead of getting dumped on with paid propaganda and disinformation so that the middle men fill their pocket.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: