ReportLegora cares deeply about hallucination — whether models make stuff up — and has built a citation system to check that what the model writes has a link. The citation system gives users a clickable link that highlights with a bounding box in the PDF (or similar) exactly where the information was taken from. The system exists because hallucinated links to courts (e.g., fabricated case citations) do happen with lawyers using AI tools.
Latent Space · Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora
Creator perspectiveBuild evals as RL-environment-style 'worlds' that mirror how lawyers actually work: projects with realistic inputs (e.g., a 10,000-file data room), a defined deliverable (a grid report), explicit red flags it must contain, and constraints like conciseness.
Latent Space · Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora
ReportJacob Lauritzen endorses using tabular review to ask the same question across all documents in a due diligence review, with one row per contract and one column per question — the agent fills out the entire thing and it's super easy to work in. Users can also just drop files into the agent and ask it to run due diligence; it will spin up tabular review, run all documents, wait for completion, then analyze results. The agent can also flag red flags in review results (e.g., unusual CEO pay or bonus structures) for lawyer follow-up.
Latent Space · Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora
ReportA typical use case for tabular review is due diligence, where a firm reviews all employment contracts, all supply contracts, and similar document sets; subsequent columns ask further questions such as whether there is an IP assignment clause, how much must be paid if the person is fired, and how much they are getting paid; the agent can highlight red flags such as a CEO getting paid an outrageous amount or a weird bonus structure; from the results the agent can draft a memo to send to the client, or do the review and then have a lawyer look at the red flags; tabular review is collaborative, so users can comment on cells for other people to look at.
Latent Space · Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora
Creator perspective · MoneyFrontier model companies have two monetization paths: sell tokens directly to everyone (consumer plus Fortune 500 with an army of FTEs), or let SaaS companies resell inference as domain experts with distribution. This is not a binary choice — labs can own the consumer chatbot market while leaving room for others to resell inference in vertical domains.
The a16z Show · HubSpot’s Co-Founder on Finding Your Next Startup Idea
Report · MoneyA survey of HubSpot customers about five or six years into the company (a couple of years before the IPO) found 20% used Salesforce.com, 20% used some other CRM, and 60% were not using a CRM at all. Many customers paying for lead generation were not doing anything with the leads because they had no CRM — which the founders found 'insane'.
hedged
The a16z Show · HubSpot’s Co-Founder on Finding Your Next Startup Idea
Report · MoneyContinuum is Runway's concept of real-time video/world models taken to the next degree, envisioned as an operating system where what streams to your screen might come from a video model. In Continuum, you can select any part of content on a website and turn it into a 'portal' — a real-time, fully interactive video based on the content you are reading. Portals can also be made through illustrations or any content the user is watching or interacting with.
hedged
AI Engineer · You Can't Tell What's Real Anymore. Here's What Comes Next - Cristóbal Valenzuela, Runway
Creator perspective · MoneyThe host shares a framework he loves: you want to be at the top coming up with direction and ideas, then AI does a bunch of stuff, and then you review it and iterate until it's done. He calls it the AI sandwich, or the human sandwich.
Lenny's Podcast · Molly Graham: The grief, burnout, and opportunity hiding inside the AI transition
Creator perspective · MindRobert Greene describes a personal morning mental practice using the image of a thin tightrope, with drama on one side and numbing out on the other. He says that getting a certain distance from both drama and numbing out turns the tightrope into a flat surface you can rest on, and that he does this practice in the early hours because algorithms and other forces can otherwise yank you down either side.
Andrew Huberman · Access the Sublime & Live Your Best Life | Robert Greene
Report · MoneyAjelix's solution to dependency and margin problems was owning its inference: it controls the entire stack including inference, host, and hardware, and can fine-tune models to customer needs while keeping them less quantized but still performing well.
MicroConf · SEO for AI Search: How Ajelix Grew to 350K Signups Without Ads
Principle · MoneyWhen you disagree with a control-plane denial decision, the first most important thing is having a way to see what decisions were taken; then escalate to a user so a human can decide whether the action should pass.
hedged
AI Engineer · A Valid Key Let an AI Agent Delete Production - Sagar Batchu, Speakeasy
Creator perspective · MoneyCoinbase sees two ways to respond to agents becoming primary actors: build and market innovative products specifically for agents that were impossible in the previous internet era, and make the core parts of Coinbase's business accessible to agents so people keep using Coinbase for trading and money management. Concretely, Coinbase is making its entire business accessible via APIs, translating all human functionality into a world agents can interact with, and building a defensive hedge to survive and thrive in the AI era — analogous to web shifting to mobile, now mobile shifting to agents.
AI Engineer · Most of Your Docs Readers Are Now AI Agents - Hahnbee Lee, Mintlify & Lincoln Murr, Coinbase
Creator perspective · MoneyPreviously, to interact with a service you had to go to the website, fund an account, get an API key, and hand it to your agent — an arduous, human-blocking process. Now, if you just give an agent a wallet or some financial account, it can pay on a one-off basis for different APIs. It's like having to walk into every grocery store, buy a gift card for that store, and sign up for its loyalty account before paying for anything.
AI Engineer · Most of Your Docs Readers Are Now AI Agents - Hahnbee Lee, Mintlify & Lincoln Murr, Coinbase
Report · MoneyWorkOS created an agent specifically for the Hilltop process called the PM, which does the first draft of the Hilltop based on the brief specification given to it. The PM agent adds context to the draft in addition to reading the human reviews, then picks up the implementation and breaks it into tickets. It solves the cold start problem — getting over the hump of creating all the different project resources. There is an opportunity for humans to stay in the loop for each part of the agent-driven process — it is very common for a human to intervene or comment on an AI-generated ticket to give further guidance or refinement.
AI Engineer · No, That's Not a Software Factory — Ryan Cooke, WorkOS
Report · MoneyMariana Minerals bought a copper mine and is reinventing the mining industry by owning the entire value chain: acquiring, developing, mining, and operating the asset, and refining the material themselves across multiple minerals, then selling a final commodity product in the market. The approach rethinks the entire end-to-end process of how you plan, operate, mine, and refine materials — not just extracting copper better.
The a16z Show · The Startups Rebuilding American Industry | The AD50
Report · MoneyManufacturing technicians themselves have the intuition and now have the software tools to change the economics of producing low-volume, high-mix on the line. A technician using these tools can reprogram and reconfigure manufacturing lines on the fly. Traditionally, retooling a line required calling a system integrator, took three months, and cost hundreds of thousands of dollars; now a manufacturing engineer can use Claude to reconfigure a manufacturing line in about 20 minutes.
The a16z Show · The Startups Rebuilding American Industry | The AD50
ReportAI companies including Anthropic and Amazon are described as buying physical books in bulk, cutting off their spines, scanning the pages for AI training data, and then destroying or shredding the originals. Joe Rogan is skeptical that destroying books is the easiest way for AI companies to get information and doesn't know that they're actually doing it, noting Amazon shredding originals is absurd since Amazon already has the books in its warehouses.
Joe Rogan expressed skepticism and uncertainty about whether this is actually happening
PowerfulJRE · Joe Rogan Experience #2571 - Jimmy Carr
Creator perspectiveJimmy Carr argues people are worried about the wrong thing with AI — not job loss but the collapse in the cost of operating an authoritarian regime. He illustrates with the Stasi, which used to cost like 20% of GDP with 30% of the population spying on the other 70%; now you can run it off a laptop.
PowerfulJRE · Joe Rogan Experience #2571 - Jimmy Carr
ReportAnthropic's internal memo for Project Panama described it as 'our effort to destructively scan all the books in the world' and advised discretion, using a codename. Joe Rogan interprets the codename and discretion as showing Anthropic knew they were doing something that would cause people to freak out and revolt.
PowerfulJRE · Joe Rogan Experience #2571 - Jimmy Carr
ReportJoe Rogan reports that Anthropic had a $1.5 billion settlement for using upwards of 7 million different books, and that in the Bartz v. Anthropic case the court decided in late July that Anthropic's training on books was fair use — they were actually allowed to do it.
hedged
PowerfulJRE · Joe Rogan Experience #2571 - Jimmy Carr