The Ledger
Every statement that passed quotation and attribution checks. Mix any filter with any other: certainty 1/5, debate potential 5/5, or both at once.
why aren't all 56 resolved? a statement only gets an assessment when the public record can support or contradict it. opinions and what-ifs never can, and 0 checkable ones are still open, waiting for their date. predictions held up or didn't; assertions are supported or contradicted. on every card: ▮▮▮▮▮ certainty · ▮▮▮▮▮ debate potential. speakers are clickable
Krieger: Top token users do not correlate with top performers
“We found that there's not a lot of correlation between
[10288] Mike Krieger: Like the person who's using the most tokens and like the person that I like, it's an interesting thought. I just made you at your companies like write down your 10 most productive peo…”
Krieger: Instagram could have reached sale with 4 to 6 people using AI
“I think we could have gotten there with like four to six, you know?”
Krieger: Anthropic found little correlation between employee token usage and actual productivity.
“We found that there's not a lot of correlation between
Like the person who's using the most tokens and like the person that I like, it's an interesting thought.
I just made you at your companies like write down your 10 most productive people that you think a…”
Krieger: AI would have allowed Instagram to scale with just 4-6 employees.
“I think we could have gotten there with like four to six, you know?”
Krieger: Anthropic dynamic workflows autonomously converted hundreds of thousands of lines of code
“We have a feature called dynamic workflows where you can have it like also break down the task into like a lot of subtasks. And I trusted it to sort of not just do the individual action, but here's like a whole language conversion of millions of hundreds of th…”
Krieger: Anthropic Labs builds for capabilities models will gain in six months
“The two most useful thought exercises we do in labs, one is like visualize the gap between what the models can do today and how most people use it. It can be closed that gap. That's one. And the other one is imagine what the models are bad at now that they're …”
Krieger: Anthropic shelved prototypes deemed potentially harmful to society
“There are certainly products that we have either prototyped or conceptualized and been like this product. It sounds so hypey. I hate this. But like this product, if shipped, would be bad for the world or like would nudge people in the wrong direction. Or even …”
Krieger: Fable completed a full night of delegated engineering in one hour.
“I moved much more to before going to bed, making sure that I had, like, queued up for Fable, like, enough chunky work to last. I would call it the whole night and I would check in later and it got it done in an hour and it was like, I guess, hanging out for th…”
Krieger: Fable autonomously converted hundreds of thousands of Python lines to TypeScript.
“And I trusted it to sort of not just do the individual action, but here's like a whole language conversion of millions of hundreds of thousands of lines of code at that point. Go off and do it, go plan it, go execute it, go verify the work, double verify the w…”
Krieger: Early prototype of Claude's computer use feature accidentally deleted user files.
“And then also out of that first incarnation came computer use. Computer use was different, though, because when we built it was really bad. Like we tried a bunch of products with it, and this was around, you know, Sonic three five and be like, Claude, can you …”
Krieger: Space-based data centers actually might make sense
“At first, I admittedly thought it was a crazy idea in general, but now I'm like, oh, I actually really can understand why this might make sense.”
Krieger: Outcome-based AI pricing will be extremely difficult for broad knowledge work.
“It gets so much fuzzier on these like tasks that we actually ask Claude these days.
Like I had a strategy document.
I use Claude to critique my strategy document.
Like,
What was the outcome?
It's like, well, I don't know.
It's like, tell me how the strat…”
Krieger: Instagram's original Android app could be built in a week with AI
“Android we built. In about a month for Instagram, we could have done it probably in a week with the models”
Krieger: AI needs 75% to 80% human quality to actually accelerate work
“If you get to, like, 50% as good as you would have done yourself, I don't think that's good enough, and it won't speed you up. And in fact, it's like, I don't know, I could have just done this myself, and now, now, then at least I would have known what it's do…”
Krieger: Claude Sonnet 4.5 beats Opus across categories at one-fifth the cost
“So, 4.5 sine 4.5 basically outdoes Opus, our largest model in effectively every category, but does so while running faster and at a fifth the cost. So if you think about where we were in May at, you know, code with Claude, we were announcing, announcing Opus f…”
Krieger: In the long run, AI products will probably replace more work
“In the long run, like overall, these products like might not, or probably will be doing more sort of automation or even replacement of work.”
Krieger: AI era will foster more small companies over monolithic firms
“I still think there's like a tremendous amount of economic opportunity throughout. It just might be, you know, more smaller companies rather than fewer, bigger monolithic companies.”
Krieger: Anthropic trained native memory directly into Claude rather than external wrappers
“Rather than treat it as a sort of substitute for how the model might otherwise access information or sort of a system built on top of the model, we actually have trained it deeply into the model. And so the model knows about the concept of memory, which I know…”
Krieger: Meta's Search Bar Chatbots Were Not Particularly Transformative
“Initial wave of, well, you know, we've got some chat bot type stuff in the search bars was like not particularly transformative.”
Krieger: Massive Platforms Can Only Introduce One New Behavior Per Generation
“Once you get a service as widespread as Instagram or as widespread as Facebook or WhatsApp, it's hard to introduce a new behavior there. You know, we did it with stories. I think they've since done it with reels but it's almost like one, you get one per genera…”
Krieger: Anthropic is embedding engineers inside enterprise clients with Deloitte
“We're doing much more of a model now where either with our own engineers embedded in enterprises in partnership with Deloitte, which we just announced this week, can we actually like take our technology, meet companies where they like what their highest needs …”
Krieger: First-week reactions to new AI models should be ignored
“I've learned to not really trust day of or even week of model reactions. You don't really know until you've put it through its paces. And so like, I almost just completely block out the noise in the first couple of days of any new model release, because I don'…”
Krieger: Anthropic is in discussions with SpaceX
“We're, you know, we're talking to SpaceX about spacey things.”
Krieger: Outcome-based pricing is very hard for general LLM work
“It gets so much fuzzier on these like tasks that we actually ask Claude these days. Like I had a strategy document. I use Claude to critique my strategy document like. What was the outcome? It's like, well, I don't know. It's like, tell me how the strategy goe…”
Krieger: Instagram's month-long Android build would take one week with AI
“In about a month for Instagram, we could have done it probably in a week with the models”
Krieger: Initial model release reactions are unreliable without real-world workflows
“I've learned to not really trust day of or even week of model reactions. You don't really know until you've put it through its paces. And so like, I almost just, Completely block out the noise in the first couple of days of any new model release, because I don…”
Krieger: Anthropic Labs designs products for model capabilities expected in six months.
“The two most useful thought exercises we do in labs, one is like visualize the gap between what the models can do today and how most people use it. It can be closed that gap. That's one. And the other one is imagine what the models are bad at now that they're …”
Krieger: Simple text-box interfaces are no longer sufficient for frontier AI
“Spending a lot of time in lab saying just a text box and a big text It's not going to cut it anymore. Like when I mentioned that the models feel like they're way smarter than me when they talk to me, sometimes I'm like, can you draw me a picture? Because this …”
Krieger: Environmental awareness and agency will transform Anthropic's entire product line.
“That's like theme one that I'm extremely excited about. And I think if we do it right, should actually like transform all of our products like from head to toe.”
Krieger: AI tools still place too much technical burden on everyday users.
“We're still asking people to understand way too many concepts of like, what is, you know, the difference between like my sandbox environment and production or like connected MCP as myself or others, or how should I store data?”
Anthropic's Claude Managed Agents offers an outcome-based mode evaluated against a rubric.
“We have a product called cloud managed agents where we'll run all of the infrastructure for you in terms of doing all of the, you know, agentic harness and
Calling the tools, et cetera.
And you can either do it in sort of the normal mode, which is you give i…”
Krieger: Data gravity prevents users from switching to 2x better AI apps.
“Even if somebody has a like two X better AI powered, you know, doc editor, you're going to move all your stuff up. Maybe, probably not.”
Krieger: Anthropic has shelved product prototypes deemed bad for the world.
“There are certainly products that we have either prototyped or conceptualized and been like this product. It sounds so hypey. I hate this. But like this product, if shipped, would be bad for the world or like would nudge people in the wrong direction. Or even …”
Krieger: Sonnet 4 and Opus 4 Lose Focus Over Long Coding Horizons
“So for example one of them was you know, Claude, you know, Sonnet four and even Opus four, Opus is our biggest model is good at writing code, but, you know, tends to get sidetracked or lost if it's working over longer time horizon. So that was a real emphasis …”
Krieger: Scaling laws describe what is possible, not what is predetermined
“The scaling laws, I think, paint a picture of what is possible, but is not predetermined. Like, to actually get there, there's a lot of, actually, really difficult, both machine learning and engineering work”
Krieger: Sonnet 4 to 4.5 gains stemmed largely from post-training engineering
“I think if I pointed at something between Sonnet four and four or five, a lot of it really has been on the engineering side to just be able to scale up especially a lot of the post-training work.”
Krieger: Claude plays a more fundamental role in operations than in code building
“I think the answer to your question is it's support. It's a supporting role on the sort of building side. But it's playing a much more fundamental role in terms of the actual operational side.”
Krieger: An Anthropic customer ran Claude 4.5 agentically for 30 hours
“We had one customer had it work for 30 hours. Of course, that's not going to be every task, but like, that's the kind of upper bound that we're starting to see.”
Krieger: Anthropic is collaborating with Microsoft on Office and Excel agents
“We've been working more closely with Microsoft, even for some agents, even within their office suite, so being able to take the financial analysis capability and the financial planning capability and bring it closer to an Excel user, for example, I think that'…”
Krieger: Knowledge workers will become AI managers rather than users
“I think people end up feeling more like managers of AI than just users of AI.”
Krieger: Top Anthropic engineers manage 3-4 Claude Code instances simultaneously
“Our best engineers are managing three or four cloud code instances running at once.”
Krieger: Social talent in AI reflects prior concentration of product talent
“I think it's less that there's a lot of social media sort of oriented folks that have now moved into AI. It's more that I think a lot of the best product people were focused on that, you know, even four years ago, you know, pre-chat GPT you know, pre the emerg…”
Krieger: Claude relies on word of mouth over viral growth loops
“With Claude, it feels quite different in that, you know, we have more of a business audience, like plenty of people use it for their individual pieces, but it has less of that sort of you know, social component right now. It's definitely more word of mouth... …”
Krieger: Anthropic tracks daily visitors for utility rather than time spent
“We don't really look at engagement, at least not in the typical, like at Instagram, we spend a lot of time looking at things like time spent. Right. We do look at things like daily visitors as a proxy for a utility.”
Krieger: Creative tools historically fail to become enduring social networks
“You would see an emergence of a creative tool and whether they were able to sort of transcend that to being a network that you come back to was often not the case.”
Krieger: Coding capability is on the critical path to agentic AI
“For us, the coding piece, beyond just the fact that coding is a really high value economic activity, I really see the model's ability to plan, write code, solve problems as not just being useful for software engineering, but being really critical path to the k…”
Krieger: Anthropic product engineering had only 25 people in 2024
“I started the labs, the original labs team with Ben Mann, who's one of the co-founders of Anthropic in my third week at Anthropic. And it had been something that I've been bubbling under. And at the time, the reason was really different. It was all of our prod…”
Anthropic assesses biological risk by measuring capability uplift for lay users
“So if you look at our model cards, for example, one of the ways we look at risk in the bio domain is comparing the uplift from sort of lay a lay person using the model versus, you know, an expert or a lay person just using the Internet and seeing what the comp…”