The weekly top 10 for B2B tech operators · Every Friday
ARR - Kyle Poyar makes a case that ARR is just not that trusty a metric for AI-native companies and proposes gross profit per million tokens (see #6 below). Late last year, Tomasz Tunguz found that this metric tracks closely with AI company trading multiples (and also revenue per employee). The logic: In an 80% margin SaaS world, ARR was a great proxy for value. In a 20-60% margin AI world (with wildly variable cost structures), it ain't.
I don't think I have more to add than the title of this article: "You Cannot Be Data-Driven Without Experimentation," but the opposite of that is also a truth to hammer home "You Cannot run experiments successfully without being data-driven". We all overestimate our experimentation skills AND our Data collection skills.
Bloody hell - The Big Book of VC is a BEAST of a PDF: AI captured 51% of all global venture funding but only 30% of deal volume - capital concentrating into fewer, bigger bets. Secondaries hit $102B in H1 2025, overtaking IPOs as an exit mechanism for the first time. Only 25% of current unicorns are realistically IPO-ready, and 60% of 2025 IPOs are priced below the last private round.
Ever wonder what the heck those ads are that have no meaningful call to action? System1 analyzed 195 US TV ads and found that ads using characters, dialogue, and expressive people drive sustained attention; product shots and voiceover lose it, even from interested buyers. For us operators: 94% of B2B buyers build their shortlist before contacting a vendor, and 77% buy from their original top choice. Being in memory before they're in-market can make you an easy part of the deal.
Bessemer's AI pricing playbook is out and identifies the core structural shift: SaaS charged for access, AI must charge for outcomes. The problem is that outcome-based pricing requires knowing your cost-to-serve per transaction before you set a price - most teams don't. Get the unit economics wrong at pricing design, and every enterprise contract becomes a margin liability at scale.
Clay just took an immediate 10% revenue hit to restructure pricing around a platform + tokens model - separating out data credits (cost pass-through at 50-90% reduction) from workflow actions (value). Kyle Poyar's read: this is where AI pricing is heading. Credit-based pricing surged 126% in 2025. The operator question is whether your customers can predict their bill, and whether your margin survives if they can't.
ChatGPT was the fastest-growing consumer app in history. Claude just became the fastest-growing enterprise product, growing from $1B to $14B ARR in 14 months. The twist: Anthropic built a one-click cancel button into the product. Lemkin's point lands harder for that reason: a 2024 FTC/ICPEN study found 76% of SaaS companies deploy dark patterns to trap churning customers. If the category winner doesn't need to, neither do you.
Folks, there is a new kind of debt in town - comprehension. An Anthropic randomized trial found that AI coding assistance produced the same velocity but 17% lower comprehension scores - and developers using AI for passive delegation scored below 40% on comprehension tests vs. above 65% for those using it for active inquiry. The tool isn't the problem; how you use it is.
Whoops! Some of Amazon's AI coding assistants contributed to four P1 incidents in one week! $6.3M in lost eCom orders and a 99% drop in order volume across North America - ouch. The response: a 90-day safety reset and mandatory two-person review for all code changes. Tomasz Tunguz's frame is the harder question: AI-generated code produces 70% more issues than human code, Utah has already eliminated the hallucination defence in law, and the company bears liability for every action its agents take.
NON-DILUTES: The VC Corner has compiled 80+ non-dilutive funding sources across grants, RBF, venture debt, prizes, and cloud credits. One founder stacked four sources to raise $1.2M at zero dilution. The stacking math is the hook: cloud credits alone (I'm a big fan) across three major providers are worth $800K+, government grant programs run 15-20% approval rates for Phase I at $50K-$500K per award, and the global RBF market hit $5.8B in 2024, growing at 70% annually.
Get the top 10, every Friday
Curated SaaS and tech insight from around the web, repackaged for people to put to good use — free.