• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Chetan Puttagunta Posts Claim on Anthropic Model Integration

    Post draws replies from engineers and researchers on model integration.

    GT
    TK
    GM
    9 Sources, 60d ago, first seen 60d ago

    TLDR

    Chetan Puttagunta wrote that Anthropic has told CEOs and investors it sees no future for separate companies selling models versus harnesses or applications. Dax Raad replied that such tools grow complex at scale and need live user data feedback. Gary Marcus called the stance a win for neurosymbolic methods. Thomas Kipf agreed the core idea holds despite the phrasing. Other replies raised concerns that vertical integration could prompt customers to seek replacements.

    Combined views

    509.7K

    9 Sources, first seen 60d ago

    Combined views

    509.7K

    9 Sources, first seen 60d ago

    2.4K likes
    2.4K likes
    233 comments
    1.1K saves
    128 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    233 comments
    1.1K saves
    128 reposts
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    9 Sources

    @chetanpIt's subtlety wrapped in technical terms that Anthropic has been broadcasting both publicly and privately to CEOs, VCs, startups. They don't see a world where the application (ie harness, etc) and the model are separate companies and therefore will compete with their customers.
    @garrytanMost interesting 2026 vibe shift is OpenAI actually looking to be the open platform Note the marked difference: intelligence on tap as a utility vs signaling it is optimal to integrate all the way up full stack
    @GaryMarcusCheckmate: you can’t take the harness (which is typically in large part symbolic) away from the neural model without giving up performance. HUGE victory for neurosymbolic AI, straight from @AnthropicAI Technical Staff.
    @mattparlmerRT @tekbog: if anthropic believes in selling model+harness - in the name of safety ofc! - then this translates into wanting to compete with…
    @evanjconradi think if anthropic really believes that they must build both the application and the model in one, this causes two problems for them: - it creates the largest possible incentive to replace them - it sets up other labs to counter position by waiting for apps to overshoot customer expectations say it's the 1920s and you've got amazing engine technology. but no matter how good your engine is, you're still going to get a faster car by building the rest of the car: the chasis, the wheels, the engine, the whole thing. but now fast forward in time and wow cars are getting really fast, even faster than the speed limit allows. but now these customers are suddenly no longer interested in the speed of the car, because every car can go 100 miles an hour. they care about leather seats. they care about radios. it's this era of the lifecycle that the modular technology shines. the maker of the car radio could instead make cars by buying the engine, a modular component, from another vendor. meanwhile, our original vertically integrated competitor may struggle to split up their products, because they make so much margin selling to the top of the market, the race car drivers who still care about speed, and because they've invested so much money vertically integrating. so a reasonably smart model vendor, looking at anthropic vertical integration play, could position themselves to take the tremendous capital that is at risk if anthropic comes for them, and "arm the rebels" a bit. in cars, this may not have worked, because the timescale from going from "20mph" to everything being "80mph" was ~100 years. but the acceleration of AI is so fast that it the vertically integrated products may quickly overshoot customer expectations, possibly within ~years, rather than decades. if i had to put myself in the mind of anthropic, i suspect their answer to this will be like "we have the newest models, and so we will always be able to use them ourselves to live in the future and outcompete everyone on products" because of the speed advantages they get. i think this framing is reasonably true, but it also means that they're exactly incentivized to push up market and attack the "race cars" of the world, which will likely earn them higher margins. but the problem of doing this is that they expose the lower end of the market to a modular competitor who either partners with someone else offering a value proposition that anthropic was not specialized in offering (the "radios in the car"), or a modular competitor who simply pushes prices down such that anthropic reallocates their hard assets towards the top of the market, where all their margins are. that modular competitor then gets to keep pushing up market at the same rate anthropic does, as a fast follower that brings down prices while anthropic (or whoever the leader is) goes off and makes the cancer cures or whatnot.
    @thdxrpeople want to dunk on this by saying their weekend harness does better on some dumb one shot benchmark but at scale this is true. these tools are complicated and must factor in user behavior. you do not even understand what this means unless you have a huge user base sending you data all of that gets sent as feedback to the model which gets trained into specific behavior i think the part that's misrepresented though is it's extremely easy to recreate the exact same environment there nothing hidden in the harness, you can observe it all and recreate it. bit tedious to keep up with it but not hard so ultimately even if they're right it's not defensible
    @tkipfThis is a good take (even though I think the “checkmate” phrasing here is really out of place): while modern model harnesses probably aren’t exactly what the neurosymbolic AI community envisioned (as otherwise they would have built them), the core tenet of combining structured representations/computations with neural models is what drives the current frontier.
    @andersonbcdefg@evanjconrad "we have the newest models, and so we will always be able to use them ourselves to live in the future and outcompete everyone on products" was easy for them to believe 6-8 months ago but now feels like pure cope/delusion

    9 Sources

    @chetanpIt's subtlety wrapped in technical terms that Anthropic has been broadcasting both publicly and privately to CEOs, VCs, startups. They don't see a world where the application (ie harness, etc) and the model are separate companies and therefore will compete with their customers.
    @garrytanMost interesting 2026 vibe shift is OpenAI actually looking to be the open platform Note the marked difference: intelligence on tap as a utility vs signaling it is optimal to integrate all the way up full stack
    @GaryMarcusCheckmate: you can’t take the harness (which is typically in large part symbolic) away from the neural model without giving up performance. HUGE victory for neurosymbolic AI, straight from @AnthropicAI Technical Staff.
    @mattparlmerRT @tekbog: if anthropic believes in selling model+harness - in the name of safety ofc! - then this translates into wanting to compete with…
    @evanjconradi think if anthropic really believes that they must build both the application and the model in one, this causes two problems for them: - it creates the largest possible incentive to replace them - it sets up other labs to counter position by waiting for apps to overshoot customer expectations say it's the 1920s and you've got amazing engine technology. but no matter how good your engine is, you're still going to get a faster car by building the rest of the car: the chasis, the wheels, the engine, the whole thing. but now fast forward in time and wow cars are getting really fast, even faster than the speed limit allows. but now these customers are suddenly no longer interested in the speed of the car, because every car can go 100 miles an hour. they care about leather seats. they care about radios. it's this era of the lifecycle that the modular technology shines. the maker of the car radio could instead make cars by buying the engine, a modular component, from another vendor. meanwhile, our original vertically integrated competitor may struggle to split up their products, because they make so much margin selling to the top of the market, the race car drivers who still care about speed, and because they've invested so much money vertically integrating. so a reasonably smart model vendor, looking at anthropic vertical integration play, could position themselves to take the tremendous capital that is at risk if anthropic comes for them, and "arm the rebels" a bit. in cars, this may not have worked, because the timescale from going from "20mph" to everything being "80mph" was ~100 years. but the acceleration of AI is so fast that it the vertically integrated products may quickly overshoot customer expectations, possibly within ~years, rather than decades. if i had to put myself in the mind of anthropic, i suspect their answer to this will be like "we have the newest models, and so we will always be able to use them ourselves to live in the future and outcompete everyone on products" because of the speed advantages they get. i think this framing is reasonably true, but it also means that they're exactly incentivized to push up market and attack the "race cars" of the world, which will likely earn them higher margins. but the problem of doing this is that they expose the lower end of the market to a modular competitor who either partners with someone else offering a value proposition that anthropic was not specialized in offering (the "radios in the car"), or a modular competitor who simply pushes prices down such that anthropic reallocates their hard assets towards the top of the market, where all their margins are. that modular competitor then gets to keep pushing up market at the same rate anthropic does, as a fast follower that brings down prices while anthropic (or whoever the leader is) goes off and makes the cancer cures or whatnot.
    @thdxrpeople want to dunk on this by saying their weekend harness does better on some dumb one shot benchmark but at scale this is true. these tools are complicated and must factor in user behavior. you do not even understand what this means unless you have a huge user base sending you data all of that gets sent as feedback to the model which gets trained into specific behavior i think the part that's misrepresented though is it's extremely easy to recreate the exact same environment there nothing hidden in the harness, you can observe it all and recreate it. bit tedious to keep up with it but not hard so ultimately even if they're right it's not defensible
    @tkipfThis is a good take (even though I think the “checkmate” phrasing here is really out of place): while modern model harnesses probably aren’t exactly what the neurosymbolic AI community envisioned (as otherwise they would have built them), the core tenet of combining structured representations/computations with neural models is what drives the current frontier.
    @andersonbcdefg@evanjconrad "we have the newest models, and so we will always be able to use them ourselves to live in the future and outcompete everyone on products" was easy for them to believe 6-8 months ago but now feels like pure cope/delusion