“Anthropic’s gross margins are above 80% before accounting for revenue shared with distribution partners, including Amazon, and the cost of training its models.” Wow.

@BreakingSaaS Sorry about that. @Gerithecfo and I were arguing about the color scheme. We were also so distracted reading about all your cool work Make sure you check out the Invoice Timing page. Here it is! app.cfo.ai/s/5ec29e80-e93b-4…

Five numbers tell you whether an AI business is a real business. Gross retention.@WeAreLegora is 95%. Customers who bought last year are still here this year. If this one is broken, nothing downstream matters. NRR. Ours is 300%+. Gross retention is the floor. NRR is how much taller customers build on top of it. We don't sell shelf-ware.  DAU/MAU. Ours is north of 50%, and the average active user spends 17 hours a month in the product. A rollout tells you a firm has signed. This indicates the work actually matters, and it moves here before it appears in retention or NRR. Win-rates. Our August pilot closed-won-win rate was 78%. Winning roughly 4 out of 5 competitive pilots is downstream from offering a superior product. Gross margin. The one that matters most. Ours is positive and improving every quarter. Our customers want Legora to be a long-term partner, and this is what makes that possible. There's a shorter route: Price below what it costs to serve, book the logo, and hop on a never-ending fundraise treadmill to pay for it. The top line goes up, everyone claps, and every new customer costs more than they pay. Scaling a negative margin only exaggerates the problem. The whole point of scale is that the margin improves as you go up. Ours does. That's the only version of this business worth building.

OpenAI subsidized for market share, then raised prices as cash tightened.

It’s time to expose a huge scam in AI startups: Contracted ARR The reason many AI startups are crushing revenue records is because they are using a dishonest metric The biggest funds in the world are supporting this and misleading journalists for PR coverage. The setup: Company signs 3-year enterprise deals. Year 1 is discounted (say $1M), Year 2 steps up ($2M), Year 3 is full price ($3M). They report $3M as “ARR” — even though they’re only collecting $1M right now. The worst part: The customer has an opt-out option at 12 months! It’s not actually a 3 year contract. In the chart below, by Q5 the company is trumpeting ~$100M “ARR” to press, while actual cash-generating, in-effect ARR is ~$35M. That’s ~3x inflation. On top of this, enterprise AI companies are bundling full-time “forward deployed engineers” into deals massively reducing margins, sometimes producing Year 1 negative margins. At some point customers are going to start triggering their opt-out clauses or aggressively negotiating down Year 3 pricing. And a wave of enterprise AI companies may collapse.

Wow, this tweet went very viral! I wanted share a possibly slightly improved version of the tweet in an "idea file". The idea of the idea file is that in this era of LLM agents, there is less of a point/need of sharing the specific code/app, you just share the idea, then the other person's agent customizes & builds it for your specific needs. So here's the idea in a gist format: gist.github.com/karpathy/442… You can give this to your agent and it can build you your own LLM wiki and guide you on how to use it etc. It's intentionally kept a little bit abstract/vague because there are so many directions to take this in. And ofc, people can adjust the idea or contribute their own in the Discussion which is cool.

LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them. IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides). Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale. Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base. Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into. Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries. Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows. TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

Claude in PowerPoint is now available in research preview for Max, Team, and Enterprise. Claude reads your layouts, fonts, and slide masters to stay on-brand — whether you're building from a template or generating a full deck from a description.

🤔