Posted by Peter216 1 day ago
I think most non-gaming apps will be replaced by a single AI app that is like a super assistant to you. No need to book on Booking.com app if you can just have your assistant look for what you actually want.
Likewise, I think Google Search is in similar amount of trouble, if not more and faster. Why do you need to search on your own if AI can search 20-30 websites in seconds and find what you're looking for?
Same for things like Microsoft's cash cow Office suites. In fact, Claude already has docs built in.
This is why I think big tech went all in on AI. All their cash cows are under huge threat by AI-first products.
Take all these basic lifestyle CRUD apps. Maybe it's a skill issue on my part, but whenever I've tried to use LLM apps over any sort of medium-to-long-term time scale for tracking something (ie. endurance running training), within about a month it begins hallucinating, recalling the wrong things, linking the wrong bits of data to other bits of data (dates, distances, heart rate data, workout paces etc), and it does all of this even when pointed to a basic spreadsheet or simple text file to keep it onboard with what happened where. It just seems to look at these files, and then look at it's own context and confuse all the details based on its own context.
Every other month, some brand new shiny model is released and it's exactly the same deal every time. I think the underlying technology simply can't deal with these sorts of goals. A whole range of apps for lifestyle tracking don't seem under threat, in my opinion, if this is the best its capable of.
Many other things I can't imagine wanting an AI to do for me, especially anything involving payments, bookings. There was a discussion the other day on here about AI being used to order takeout. Why would somebody want to delegate one of their last personal decisions to a LLM?
Harnesses have been developed for this, and its also fairly trivial to do large scale verifiable-rewards post-training for "personal assistant" style things. eg., invent a booking website, invent a PA GUI-aware harness, and reward the model when a successful booking takes place. Do that 1bn times.
Is a "recent orders" or "reorder" feature not enough to meet this demand if you frequently order the same thing from the same place? I don't quite understand the need for a middleman LLM here.
When you're asking for suggestion, is that a completely random suggestion to try new dishes or a pre-selected range to randomise from a choice of things you've already pre-screened?
UberEats should have a MCP service that is discoverable by agents - lets call her Siri. I tell Siri what I want, and Siri is checking via that MCP API what restaurants, what type of food, prices, ratings etc. Then Siri is free to generate whatever UI is needed to present the order. This could be generated HTML, but also be read back to me with voice, or even displayed on my Apple TV if I was just watching a series.
I am a professional frontend/UI dev working as a freelancer, and I myself think there will be no more market for creating UIs for the product - the UI is something that will be generated on the fly by the LLM agent for the end-user in what ever way and context the end-user specified. All you need is access to the data and services - MCP just entered the room.
However, about 1 year ago, AI coding agents were laughed. Now the vast majority of code is written by them.
Intelligence getting exponentially cheaper per token. At some point, there will be an inflection point where "it just works".
I may be completely out of the loop here, so forgive me if I'm missing details, but I always have and continue to believe that the coding prowess of LLM agents more-or-less comes down to the sheer abundance of data it has trained off of on the web, and that code itself is ideal for LLM by the way it is discrete and syntax-bound, same as the way LLM can write very well provided they've read more-or-less the same sequence of text before in abundance. For the speed advantages it provides in coding, and for the accuracy it has for well-established information, LLM is a very useful technology.
I just don't follow how they are going to do the same thing with such granular telemetry data, such as datasets from a day of nutrition or a workout. The values are continuous, minute, and are so parallel to noise in the way they are provided. Take running. If you dumped a .fit file from a Garmin or other watch with a series of heart rate values, paces, elevation levels, GPS co-ordinates, respiration rate, vertical oscillation etc, and you trained on a bunch of files like this, the values are so discrete that I can't understand how it would connect them and tokenize them across different people or even the same person's various workouts, with the high degree of factors (nutrition, climate, illness/injury state and severity) that can alter those values. It's no longer about just tokenizing text, but with advanced calculations.
I've seen second-hand through friends in running with apps like Runna that it produces plans that aren't suited for them at all. There's simply too many small variables that these things don't seem to be able to map. It gets caught up on summaries of runs. It can give general advice about things, but it doesn't have the level of detail or understanding necessary to give good guidance. Strava "AI" and Garmin "AI" just make a 2-3 line summary from the average data from the entire run or workout.
Perhaps this is something that is already being worked on, and I would be happy if it was, but I just can't personally see how LLM can work with such datasets in a detailed way, beyond the useless summaries Garmin and Strava give "You ran 10km in 35:00, with a pace of 3:30/km. This shows that your speed is improving". I would love an explanation/links on how LLM can get around this if anybody is willing to spend the time providing it. Happy to have my mind changed.
The problem with these Garmin and Strava AI summaries is that most of them just point an LLM at a heap of data and ask them to offer a summary. They also usually use older models since the feature is not something particularly well curated. I do financial modeling with LLMs and use the LLMs to build/verify/iterate on models.
If you're curious on what it takes to run long-running tasks Anthropic has https://www.anthropic.com/research/harness-design-long-runni... out.
For example, I use LLMs to create 3d models. I don't tell them to recreate the 3d vertexes every time, they have a python script that programmatically creates the shapes, and with a freecad mcp, they get views in images, also they can create other scripts to check for collisions and other issues.
They can also get a histogram, discard noisy values, check outliers, etc, etc, using scripts or by running a traditional ML value to get later the refined data.
Opus 5.5 is quite decent currently, but expensive and not enough for many challenges, but I see the progress happening.
Is that really true outside of the HN bubble? I'd like to see some studies that back it up.
And there, it has a database of sorts, running god knows where on god knows what, and you have a single-user web app that you may or may not ever open and see (since your favorite AI agent can just read and update it for you anyway).
Today it’s still something the user has to suggest / request.
Tomorrow? If they put a bit of effort these models will reach for building these one-off ephemeral or long-lived hyper-custom apps the same way they now reach for a python sandbox for basic arithmetics and statistics.
Looking for a flat? Scheduled task + CRM like artefact for it to monitor, maintain, and update.
Looking for holidays? Your own personal shareable moodboard with destinations and what not.
Have a use case or task for which chat or editing text docs is a frustrating way of correcting or giving feedback? Have it build a dedicated, custom artefact just for that.
When you encounter a use case where LLMs limits show, make them build tools.
I wonder how open these services will be to data storage if they do become a "universal app". If I used LLM to make an app for tracking training for endurance running, and I slapped an 800kb .fit file in there each day, the storage for all of these files would probably be fine, but without condensing each run down to a textual summary like Garmin/Strava currently does ("You ran 10km in 35:00, at a pace of 3:30/km. This improved your speed/{vague motivational LLM quote}"), wouldn't the context become enormous after a pretty short duration, causing huge costs? Genuinely curious.
a) I won't trust booking's own assistant or "AI" and it will be garbage if/when they introduce it (despite fact that in order to handle it correcly it would need to have amount of information about me and my family and my trip that im not comfortable to provide them)
b) They will block 3rd party/independent assistant via CF or other measures - they want you to be on their property browsing on their terms upselling you features, not access via API/MCP
c) chat interface and general UI clunkiness related to AI and this kind of use case - when browsing booking I'm able to narrow to offers I want quickly, where accommodations are located on the map etc, lol, just thinking about doing all of that via chat UI and even more via voice interface makes me tired
That said, I don't know about most of you but booking a hotel / flight is the kind of thing I do maybe once a year - it's too incidental (for me) to even consider automating.
That said, Google (search) is in trouble, although they're ready for it because their own AI products are solid and they already have the infrastructure and whatnot to run them, unlike e.g. Anthropic or OpenAI for who their AI products are their whole business. Likewise, Q&A sites like Stack Overflow, Quora are at risk.
Even Wikipedia is seeing less visitors, but I'm not too worried about that because visitor count isn't what they need to (or should) optimize for - their main risk is articles getting poisoned by LLM generated edits.
Browsing through hotels is one of the best parts of the entire experience.
I prefer the part where you wait in airport or on train station over it.
Not saying I like being stuck at airport, but that is less annoying that selecting accommodations.
People think it's some kind of genius move or very Apple-like move to not join the AI race.
But who is actually going to build apps on the fly better, faster for you? Siri or Opus 9? Siri or GPT 10 Astra?
When many tasks are done better on a single AI app, the iPhone moat will be greatly diminished which will open up competitors to iOS and Android.
There's really no good reason I should have found it easier to ask Claude to make me a general purpose card game engine with a bunch of game rules baked in than to be able to actually find the first non-enshittified app, but thanks to all of that nonsense it was in fact both faster and better to ask Claude. And this without even needing a paid Claude account.
I wouldn't necessarily trust it with e.g. a vibe-coded Duolingo yet, but there's some other, smaller, language apps it can certainly produce at this point, I know because I did some of them last year. Same for turning my Raspberry Pi into a webcam: I'm sure there was already software for it, but asking the AI to do it a year ago was faster and easier than finding it.
And the way things keep speeding up, today's "over a month of iteration and corrective feedback" is next year's "on the fly".
I paste a screenshot and ChatGPT automatically knows to extract the text for me. Needed an expensive/inconsistent app before.
There never was any need for these apps. A website works perfectly fine for simple use cases like that.