Parse notes
Sponsored
Ads get ignored on social media but not in TLDR! Reach founders, operators and startup leaders where they actually pay attention. Learn more about sponsorship opportunities [4].
AI labs are in a price war over tokens while the GPUs behind those tokens cost more to rent. Labs try to secure GPUs on multiyear contracts at $2 to $4 an hour, then pay up to $13 once that capacity runs out. An investor's analysis estimates that Anthropic's average GPU cost rose 82% in two quarters. Anthropic's $5 billion short-term compute bill in Q3 helped turn an operating profit back into a loss. The analysis predicts labs will push enterprise buyers into multiyear minimum commitments so their customer contracts run as long as their GPU leases.
AI spend peaked in July. The top 1% of firms on Ramp spent less on AI per employee in September than in July. Anthropic added $5.4 billion in revenue in the third quarter, down from $6.8 billion in Q2. Adoption across all other firms is still rising.
Billing involves several distinct jobs. Teams without billing expertise risk making bad choices with lasting consequences. AI companies should own usage measurement, pricing logic, and customer balances, but use specialist software for invoicing and revenue recognition. The back office should own a shared model of customer agreements.
Calling a product a team of agents can make it sound more useful even when the features haven't changed. Irrational Labs showed 1,014 US software buyers the same product page with different agent labels and asked whether they'd buy it and what they'd pay. The team label raised stated purchase interest by 6.3 percentage points over a single agent, but it didn't significantly beat wording with no agent and didn't raise willingness to pay. Buyers also rated the team version harder to use, so the researchers recommend making each agent's separate job visible and keeping setup clear.
A simple sales sheet could sit in Firecrawl's design queue for three or four days because the designer had other priorities. One of the company's product marketers built Alloy, an internal workspace that gives AI agents Firecrawl's brand rules and past approved work for each new asset. Editable product demos have been the biggest win, since changing wording or timing before launch no longer means recording the whole sequence again. The team has 63 reusable animation packages and says Alloy has saved days of work. Experiments stay on a canvas until someone deliberately publishes them for teammates.
Claude Enterprise Managed Auth lets organizations connect their users to MCP servers without sending each other through an OAuth consent screen. Claude presents authorization servers with an identity assertion and receives an access token after the assertion is validated. Existing access tokens remain valid until their specified lifetimes, even if a user is removed from the organization.
Zapier Zaps now fix themselves when they break. An agent reads the error messages, works out a fix, tests it, and publishes a new version of the Zap. Changes are emailed to users. Small fixes will go through on their own - the agent will ask users before any big changes.
Track visiting AI agents and configure individual access policies for websites.
Competitors used to wait until your product was obviously winning before they copied it. SuperMe's CEO watched rivals copy Grubhub and Pinterest years ago and says four venture-backed startups copied SuperMe in recent months, down to its navigation and custom icons. Copying is now cheap enough that rivals imitate a product before they know whether it's worth copying, so the CEO puts more weight on distribution and networks that keep growing once the products look alike. Grub Canada once copied Grubhub's HTML outright but couldn't copy the restaurant network behind the site.
Agents are starting to use the same software your team does, and their needs can pull a product in a new direction. Notion's CEO says agents now create roughly half of Notion's pages, and MCP traffic from connected agents has grown about 60 times this year. He is steering Notion toward keeping company context up to date for other companies' agents as well as its own. Notion also moved finance reviews from monthly to several times a week to keep up with model releases.
An agent with access to every Slack thread and call transcript at your company can still miss what you need done today. Linear's cofounder argues that storing more information still leaves someone to decide what the work is for and what a good result looks like. A useful briefing is short because someone chose what to leave out, so he recommends explaining why decisions were made and what the agent shouldn't change. Those answers need to stay current, since a six-month-old explanation can still look clear and credible after the team has changed direction.
AI lowers the barrier to entry for coding, but it doesn't remove it.
Notion is behind the billboard with a row of app icons floating over the moon and no tagline.
An agent's summary of finished work can hide requirements it reinterpreted, so ask it to list the decisions it made for you.