It's going to be very tempting to give tool access to these agents because that is the main way they become useful, but the security model of that is challenging. Every day users will have a hard time understanding the security risk of decisions.
What is the state of "voice-print" recognition? At what length is it meaningful biometric data?
As meetings increasingly get passed through LLMs, it seems like any of these companies can just keep a voice-print vector for matching on anyone they've ever heard.
Most voices are pretty recognizable, but I don't know if this is a one in a million thing, or a one in ten thousand.
Slightly off-topic: How do yo feel Opus 5.5 can be even better. Like with Fable 5.2 it’s tough to notice the difference in my line of work and what mattered to my business I’m not even sure Fable was better.
I then ended up with the question, what more could i wish for from a model? I feel like all coding work is done now. I couldnt come up with more stuff to do. L
A model that does even fewer Claude-isms (Opus 5 was horrible for that, 5.5 is better), can also read the room better (if I ask it to write some delivery docs it still sometimes overshares internal details rather than sticking with the format and style of the previous 20 versions even after AGENTS.md tells it not to vomit internal implementation details there, though honestly something like Sol 6.1 was even worse and couldn’t even follow the Markdown template properly some days, inventing new sections) and also would have increasingly better capabilities - 3D modelling is still subpar, reasoning about images still misses details (e.g. while working on texturing fails to notice a red colored band due to bad unwrapping) and also wouldn’t get unnecessarily held back, like reverse engineering software to realize why it doesn’t work or their overall opposition to REA. Aside from that, maybe also not Claude slop UI styling with those fucking disgusting punchy short sentence labels everywhere and also oversharing details nobody cares about.
It felt humbling to be releasing a BYOK meeting Zoom app that does something very close to this on the same day they announced it! But I like our spin on it better. https://getqoral.com/bwat
Surprised otter and others are partnering. This looks like a shot across the bow for them.
Most orgs dropped them when AI got infused in the meeting apps themselves I thought
Why on earth would any third party continue to use Anthropic’s offerings to manage their AI needs? They clearly are going to cannibalize any profitable integrator.
This is something I wonder about as well. If this is their strategy then using anthropic to help you write software, directly as a feature of your product, or their office software is just giving away trade secrets/road maps/etc to a competitor.
Specifically I wonder how Google feels leasing TPUs to Anthropic while they're clearly developing enterprise office software to directly compete with Google.
Probably the same way Google feels paying billions to Apple for search traffic while Apple develops iOS, and the same way Apple feels paying billions to Google for models and compute to power Siri AI while Google develops Android, and the same way Netflix feels running everything on AWS while Amazon develops Prime Video, and...
The tech industry is highly interdependent. Each of these deals has been structured to be as mutually beneficial as possible - otherwise, the two parties wouldn't have agreed.
Anthropic have thrown a lot at the wall and yet they’re still not leading in any software, even Claude Code has been usurped by endless examples of third-party harnesses / editors / IDEs. Anthropic continue to prove that deep expertise + obsession when building software leads to better products.
> Why on earth would any third party continue to use Anthropic’s offerings to manage their AI needs? They clearly are going to cannibalize any profitable integrator.
Because their competitors will. In other words company A wants to get there faster than a company B in the same market. If there are no competitors and it's monopoly and it's only company A vs Anthropic, then yeah they should think twice perhaps before sending their code to Anthropic (or anyone else like that).
So short term you need to use their products to keep up with competitors, but longer term they effectively replace your companies ability to prove a value add.
Seems like its lose-lose. But theres no alternative in the short term for you.
I know that companies are selling data of how there white-collar worked to AI for a lot of money. The founder gets the money, but whilst he thinks he’s replacing his employees he might be also replacing his company.
I don’t know how this all plays out. FWIW I’ve turned off the data usage agreement and probs everybody should and be a bit more closed generally :/.
I kinda don't see how that's relevant? Anthropic doesn't need Granola or any other such tool to be using Claude in order for Anthropic to know that it's a product worth copying. But I guess my point then pushes things towards "Anthropic is able to copy you even if you don't use their service", which could be said for any AI lab. Just look at Muse following the hype around Grok Bot or Instinct. I'll leave it as an exercise for the reader to find more examples.
I think it's also an argument that no private labs should've developed the underlying models. It's been wreckless, based on theft of work, and now creates a power imbalance we haven't seen in a long time if ever.
If we could do it all over again we should've made it a public private development where the cutting edge research and models were developed as a public good and let private actors develop it for business and other offerings. Things would undoubtedly moved slower, but I think that would have been a benefit as well.
Because they're not spying on your meetings or your code to do it -- it happens regardless, if they take the time.
Also, because anyone who's worried about their company getting out competeted by Anthropic hasn't looked up anyway, so you can't expect much in regards to their longterm strategy. That's "iPhone Moment" thinking; completely insufficient for the gravity of 2027.
It'll be like bringing a translator. Your AIs have your own project context loaded. Except, they are really probably more knowledgeable than the humans who haven't read any of the relevant documents for the meeting yet.
Except Claude will bs in the same way as a human about having read everything and when called out will just partially it "You're right. I did overstate how much I context I had read. Page 14 was beyond the context of what I had read. Would you like me to read all of the project now?" [pinky promise this time. Spoiler: it'll lie again]
But why is one company trying to do everything? Why do the intelligence vendors need to go into the app layer? Why did the go in to the coding tool layer? They could have let a rich ecosystem build around the API but they are killing start ups all over the place. Entire categories won’t get investment or traction because they are actively trying to be a monopoly
Your dream is to be stuck on a Zoom call with an agent telling you “you’re absolutely right”?
I suppose perhaps I’m being unfair. Maybe your dream is for you and all other workers being replaced by an AI entirely. I’m sure no unforeseen negative consequences could arise from such an outcome.
Everyone knows your rug pull kind of failed by missing the IPO target 40% short. Just stop it with these half baked product releases no one cares about and fix Claude Code.
As meetings increasingly get passed through LLMs, it seems like any of these companies can just keep a voice-print vector for matching on anyone they've ever heard.
Most voices are pretty recognizable, but I don't know if this is a one in a million thing, or a one in ten thousand.
I then ended up with the question, what more could i wish for from a model? I feel like all coding work is done now. I couldnt come up with more stuff to do. L
Specifically I wonder how Google feels leasing TPUs to Anthropic while they're clearly developing enterprise office software to directly compete with Google.
The tech industry is highly interdependent. Each of these deals has been structured to be as mutually beneficial as possible - otherwise, the two parties wouldn't have agreed.
Because their competitors will. In other words company A wants to get there faster than a company B in the same market. If there are no competitors and it's monopoly and it's only company A vs Anthropic, then yeah they should think twice perhaps before sending their code to Anthropic (or anyone else like that).
Seems like its lose-lose. But theres no alternative in the short term for you.
I don’t know how this all plays out. FWIW I’ve turned off the data usage agreement and probs everybody should and be a bit more closed generally :/.
If we could do it all over again we should've made it a public private development where the cutting edge research and models were developed as a public good and let private actors develop it for business and other offerings. Things would undoubtedly moved slower, but I think that would have been a benefit as well.
Also, because anyone who's worried about their company getting out competeted by Anthropic hasn't looked up anyway, so you can't expect much in regards to their longterm strategy. That's "iPhone Moment" thinking; completely insufficient for the gravity of 2027.
Will external contacts join calls with their own Claudes?
Are Claudes going to switch to a non speech audio protocol when humans leave the channel?
Are they going to communicate with ultrasound when humans are present?
But we are quickly moving away from that point.
Like of course the ai is in on your calls and you can just ask it to do stuff… anything less wouldn’t be sci-fi cool
AI is displacing jobs faster than it is creating them. A Claude agent in a meeting doesn’t really mean much if you have no meeting to attend
I suppose perhaps I’m being unfair. Maybe your dream is for you and all other workers being replaced by an AI entirely. I’m sure no unforeseen negative consequences could arise from such an outcome.
Wiring an LLM into elevenlabs and having it sit amidst humans sending each other voice memos is great. I don't need a closed-source solution for that.
We do it here:
https://magenta.cryptograss.live/