Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO. Read more: anthropic.com/news/s1-sec Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.

Most investors are still massively undervaluing data. I was also skeptical when I left a data co for Altos in 2025. Today, I’m all in on data. Every software company will become a data company. Most will lose. Many more will win than we think.

Technical moats are mostly over at this point. If you’re building something that is meant to be durable, the only moats now are: • Proprietary data • Network effects • A physical component

Mercor Research mapped AI benchmarks to actual US jobs. Only 208 of 923 occupations in the US economy are represented by existing AI benchmarks. 90% of hours worked in the US aren't covered well by a relevant AI benchmark. Today, we’re releasing the 𝗕𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸 𝗠𝗮𝗽 𝗼𝗳 𝘁𝗵𝗲 𝗨𝗦 𝗘𝗰𝗼𝗻𝗼𝗺𝘆, an interactive breakdown of which economic tasks are covered by AI benchmarks. Benchmarks are how we measure AI’s progress and anticipate its impact on real work. They’re crucial to understand which jobs AI can automate, where the biggest gaps are, and how those capabilities are changing over time. The Benchmark Map is our first step toward building a resource that makes it easier to answer: 𝘋𝘰𝘦𝘴 𝘵𝘩𝘪𝘴 𝘈𝘐 𝘮𝘰𝘥𝘦𝘭 𝘸𝘰𝘳𝘬 𝘧𝘰𝘳 𝘮𝘺 𝘰𝘤𝘤𝘶𝘱𝘢𝘵𝘪𝘰𝘯? Explore it here: mercor.com/research/labor-in…

it is obvious that the data broker space is overcrowded. I would not recommend becoming a general data broker nowadays. with that said, here are ACTUAL more niche ways to earn money in the data space now: 1. de-identification. we spend a stupefying amount on on it. around 35k/TB on non-edge cases! this is at least a 7 figure monthly cost for brokers like us. it is a pricey tax that must be paid. there’s relatively few companies that do de-id well. run Presidio and GLiNER and maybe a 32b Qwen model for hard cases. then run and fine-tune based off of results. even that simple pilot could show signs of pareto! it is crucial your de-id system needs to: a) rigorously sanitise to industry standards (obviously) b) map multiple variables to a single entity (e.g erik newsham, E.N, erok gets translated into a de-id representation of me) 2. devote yourself to the buy side. since everyone and their grandma has become a broker, there is now a glut of brokers with company data to sell, yet they have no one to sell it to! devote yourself to finding data buyers. a few efficient ways to increase surface area of buyers is to: go to SF and get face to face with researchers, mass cold email outbound or fill out our grepped partners application as we are actively buying. once you have found data acquirers, tell brokers you can help them offload for 20% commission. 3. build a financial market for corpora. the current data market is unlit. often two labs bidding for the same corpus have given us vastly different prices (which is not automatically bad as there isn’t one fixed value for data). we have been asked directly by several labs for historical transaction data. as data is so fuzzy and context-dependent it would obviously be hard to standardize, so maybe the form this market would take is a marketplace. a safe air-gapped agentic system could scour an entire corpus, verify, note and map everything needed for a buyer to make an informed decision. solving data's Arrow Information Paradox. a corpus could be sectioned and sold as parts. overtime, transactional data is produced and “data class indexes” could emerge and efficiency in this currently inefficient market would flourish. companies could list data as an asset and “IPO” their corpora. there are obviously many variables (license time and exclusivity for example) but through a streamlined process that irons out rights, QC, de-id and other commonplace wrinkles. this would be more like art indexes than crude indexes 4. make autonomous lab machinery. I am super super autonomous lab pilled. ask claude to cure cancer and it will come back with experiments that need to be done to further advancement. we simply have not conducted enough experiments. the (very) simplified loop of an autonomous lab is goal→conduct millions of simulated experiments→select top performers (and a few moonshots) for real world experiment→use successes and failures to update simulation and make progress. periodic labs and anthropics wet lab are really cool examples of this. from what I have heard though, the machinery isn’t super great and is a major bottleneck. 5. broker financing. some companies would rather $400k now than $600k in a month. for an instant deal to happen, a broker needs money and a lot of these new brokers have, uhh, interestingly sized liquidity pools (they are broke). by lending them money to buy corpora upfront, they close a deal that would never have happened and you make a nice slice of the spread for providing them cash short term.

a few of my predictions for the next 6 months of ai data: - a monopolistic human data buy side layer emerges. right now you can use zap mail, heyreach, instantly and claude max plan to reach out to practically every potential human data buyer on the world in 7 days and (even with the worst strategy) at least come out with roughly 20 buyers spanning rl builders, neo labs and (if lucky) one or two major labs, who you could easily offload $20-$50m cumulatively from the very start. it is not going to be this easy to find buyers forever. as labs inevitably get f*cked by bad apples on rights, de-id, QC and the rest, a select few will emerge as “transaction gates” in which every corpora must flow them in order for the labs to deem as safe to buy. analogous to how a few labs already don’t touch corpora if it was de-id by Tonic Textual. - the waterline continues to rise. ai labs are buying real company data to build artificial workspaces to teach their models how to do an office job. using a headcount as a metric, 3 months ago they bought company internals with 10+ headcount, now they almost exclusively purchase from companies with 50 or more employees. as models saturate these smaller workspaces, demand for smaller company corpora plummets whilst demand for the already dwindling supply of large sprawling company internals increases. yet doesn’t skyrocket as at a certain point if the price becomes too high then labs fall back on synthetic expert work. it goes without saying that small companies in a esoteric domain models underperform in that hold vertical depth (e.g a 30-man freight brokerage) will still hold immense value. - reality gets graded more and more. as the most knowledgeable domain experts begin to fail in writing valid rubrics that aren’t instantly saturated we are going to have to get reality to play the verifier, domain expert and environment, if we want progress to continue. rubrics shift from the attributes of a perfect NDA to a lab measured critical tempurature. - atom environments. why hasn’t the vertiginous increase of ai capabilities led to huge economic growth? how can we get non-tech people to understand how intelligent ai is already? how do we find the best value/token ratio? to “solve” these questions I believe need to give ai a “body” in which it can interact with the real world at extremely low bandwidth. giving ai a body sounds like a humanoid which is obviously just one small slice of real world interaction, a little bit like how literally none of the ways we interact with in the real world is only via our real body but via tools instead like handles, vehicles, instruments and cutlery. maybe I should reframe it from giving ai a body to giving the world ai - protocols for company data emerge. for such a fuzzy yet crucial area of ai, it is insane how unorganized certain steps in company data brokerage are, such as giving context on what your companies internals actually hold.

An expert corrects a mistake. What happens on the next attempt? At Comprehension Labs, we're interested in how feedback changes work and whether that change carries into a different problem. Take a code review. An engineer submits an implementation. A reviewer identifies an edge case, explains the issue and requests a change. The engineer revises the code and adds a test. Together, the task context, original version, review, correction and result form a production trace with human judgment attached. Now connect that record to the engineer's next task. Do they catch a similar issue before review? Can they apply the feedback when the requirements change? Does the same mistake recur? That connection is the longitudinal part. It lets us examine how decisions and approaches change across repeated work. A successful edit shows that a correction was made. Later attempts give us evidence to investigate whether the feedback carried forward. For that evidence to be useful, the details need to stay connected. Feedback belongs with the version it addressed. An approval needs its criteria. An outcome needs its context. Expert checks and recorded explanations help make those decisions interpretable. This is what we're building around: production traces organized into connected histories of attempts, feedback, revisions and outcomes. Our goal is to make these records useful for AI training and evaluation while preserving the context that gives them meaning. If you're buying or collecting this kind of data, or your company captures these traces in its workflows, we'd love to chat. DM us or book a call at cal.com/comprehensionlabs/30…