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 22 resolved? a statement only gets an assessment when the public record can support or contradict it. opinions and what-ifs never can, and 1 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
Bialecki: Most businesses will deploy their own AI agent by year-end
“By the end of this year, I think most businesses are gonna have their own agent. They deploy either on their website or behind a phone number or through email or to chat with your Claude that you can just access.”
Lemkin: Custom AI tools outperform $50k to $80k commercial SaaS solutions
“Is that listen, you could buy this functionality exists in other products, or you could buy it for 50 grand or 80 grand a year today, right? But because it is a well-known body of software, Claude knows it. So Claude can just build it for you. Buy, don't build…”
Lemkin: Running Claude over Replit via MCP forces cleaner code completion
“These models are goal-seeking, right? And they want to finish projects. So Replit just wants, Amelia, you know this, Replit wants to finish. Replit wants to finish, and so does CloudCode if you use it on an own. But when, but if you use Cloud on top of it coun…”
Lemkin: Claude's MCP capabilities threaten dedicated agent orchestration startups
“It is a threat to folks that are building these sort of agent substrates, which is that we, listen, every time we talk about our agents and stuff, a bunch of folks will tweet back or say, Hey, listen, I've got the substrate. I've got the, I've got a way to con…”
Masad: Replit users were six months ahead of AI labs on agents
“In reality, actually, Reply users were like six months ahead of everyone else, even people in the labs, because when we launch these agents, a lot of times we get researchers messaging us, it's like, oh, we didn't know Claude was capable of running for, you kn…”
Masad: Replit's memory compaction algorithm beats Claude's
“We actually think ours is much better than Claude and many others on the market.”
Lemkin: Implementing custom OAuth in vibe coding creates major security leaks
“All of these apps have their own OAuth built in. That is secure. That has been hardened. And everyone goes in and I know Replit the best because that's where I've spent time, but they all have the same ultimate. They're all more similar than they're different.…”
Glean CPO: External AI harnesses grow faster than native Glean UI
“I mean, if I bundle all of the cloud, cursor, codecs, it is one of the fastest growing parts and definitely faster growing than the UI of Glean, which is still by a large margin, a very high daily engagement driver.”
Ibarra: SaaStr's AI agent autonomously evaluated and selected Microsoft Clarity
“You should add Microsoft Clarity. It's a heat mapping tool. It's free. It literally picked this vendor for me, and it, to be fair, it saw we had Vector, and it said you could use those ones too if you want to switch, but I assume you want to use the one that y…”
Dorfman: Anthropic sales forecasts are largely run by Claude
“What's most important is forecasts are being largely run by Claude and then inspected and reviewed by managers.”
Lemkin: AI-vibe-coded sites all look like Claude and share a recognizable smell
“At the end of the day, all these sites look like Claude. Once you see one Replit or Lovable site and you can just smell it. You know, there, someone was saying that like, 30% of this YC class vibe coded their app. And I want to see a couple. I'm like, I, that …”
Amble: Fine-Tuned GPT-4o Reached 90%+ Accuracy vs 20% Base in Accounting
“I met an accounting AI company a couple of weeks ago that had said, They tested GPT-IV-O against their benchmarks, and it was like, 20% accurate. They then tested Claude, and it was like, 50 to 80%, and then a fine-tuned version of GPT-IV-O, they were getting …”
Ibarra: Layering ZoomInfo and Clay in Cowork yielded best enrichment data
“So it used ZoomInfo in clay in cowork, three layers deep of agents to enrich this list. And it was like the best enrichment we've gotten so far.”
Salyers: Connected AI agents resolved SaaStr's five-year pension tracking issue
“When they started talking to each other, they just reconciled all the data. They looked at all of our past history of how much, you know, we've put in and they came up with a recommendation, which is all I ever wanted. And so five years for five years.”
Salyers: Claude offers far more native connectors than Replit
“Claude has a lot more connectors than Replit does, like native, right? Like, Replit, you have to build a lot of connectors.”
Anthropic encoded top rep behaviors into five Claude skills for AEs
“And we built a sales plugin that documented what our top reps do and turned them into five skills that we included in Claude that every rep uses every single day.”
Lemkin: Auditing AI outreach requires asking if you would buy from it
“When you're reading the output of an agent, you don't, you can't just ask yourself actually anymore if it's good because it just might be good. In fact, Claude keeps getting better. So by the end of the year, the output might be borderline great. Okay. You hav…”
Lemkin: Goal-seeking AI agents cut corners by doing minimal work
“Agents are goal-seeking, right? Most of them under the hood are clawed. It'll be more, we use Gemini ourselves too. We'll use more open ad. Most of them are clawed, but it doesn't matter. They're goal-seeking. And that still creates a certain laziness, which i…”
Lemkin: Replit deploys an OpenAI-powered Architect sub-agent for big features
“We actually get an extra bonus model because when Claude on Opus talks to Replit on Sonnet, if it's a big feature, Replit brings in a sub-agent called the Architect. And the Architect, it turns out, runs on Codex and OpenAI.”
Anthropic uses Claude via Slack to triage internal sales support tickets
“So we made Slack the front door. Slack comes in, tick it out. Claude does the triaging. So Claude actually co-work now does a lot of this. And that's changed for us probably in the last week since I made these slides, but co-work can check your emails and can …”
Anthropic uses Claude to surface six dynamic weekly coaching moments
“Every week Claude surfaces six coaching moments. And this isn't a static set of coaching based on a methodology. This is dynamic based on how the needs of our business evolve month over month.”
Anthropic uses Claude in Slack to answer employee questions and onboard staff
“Across both the technical and non-technical org, one of the favorite use cases is the Slack channel that we spun up, and what this Slack channel does is employees can go in, ask questions. Claude is on the back end and uses to go search over our internal knowl…”