Weekly Links: eCommerce & AI, Economics, and Salesforce Buys Fin

This week: Anthropic's Fable is still offline, SpaceX IPOs, and we got some insights into the Elias Thorne mystery.

I'm also a little sad because Midjourney's model shifts are making it hard to maintain the robot character I've been using. So far, I haven't been able to recreate the exact same look by pinning. Working on it.

On to the main stories this week:

• The 11 standout startups from YC’s Demo Day, according to VCs. TechCrunch picks out its favorite Y-combinator companies from the latest batch. It's interesting to see the diversity of companies, but also hard to unsee that all but one are AI-related. This is unsurprising since it's the technology that's driving the most change, but for some of the applications, things are moving so fast it may be hard for the startups to truly reach some escape velocity.
• The first-ever robotic rescue at sea is a milestone. An autonomous sea drone rescued an Apache Helicopter crew in the Strait of Hormuz in early June. Such events will likely become routine as the number of automated battlefield systems increases, but this was a first when it happened. Perhaps there will be another transition further down the line when there won't be any human crews on helicopters in the first place. 
• Meet the world’s top AI-pilled economists.  The real conclusion of this article is that there are few, and that the amount of economics research being done to assess the potential impact of AI is much less than would be warranted by the potential impact. Academic researchers seem, for some reason, not to be grappling much with the topic, government statistics offices are trying but have limited resources, and independents are mostly working at the frontier model labs. Many of the “economic impact” studies to date seem woefully inadequate, bordering on anecdotal. Studies from the labs show increases in productivity (which is obvious, but it’s unclear how far this can spread without hitting other bottlenecks). Most other studies fail to look at how AI will alter workflows entirely or even eliminate the need for certain activities. Personally, I believe the productivity gains will be enormous, and there are whole areas of the economy that will be rewired over the next 5-10 years. It just likely won't be by improving every job by 20%; it will be when AI can be used to collapse a whole set of tasks to zero or extremely low cost reliably. It seems critical that more studies be designed and more theories developed about the potential impact. 
• Salesforce Signs Definitive Agreement to Acquire Fin. Salesforce will acquire Fin for approximately $3.6 billion. Fin started life as Intercom: a beloved early customer support tool that embedded directly into SaaS software. The founder-led company has been one of the poster children of how to lean into the AI Agent revolution and has gone all in to build new LLM-powered customer support experiences with its Fin Agent. Interestingly, Salesforce has its own Customer Support bots as part of its Agentforce suite and has been betting heavily on these to offset widely expected SaaS revenue slowdowns. The biggest message here is likely that they really needed the team behind Fin to try to lift their entire agent suite.
• Fable 5 ban: 4 open models responded before Anthropic could restore access. In general, I love NewStack’s coverage, but I think this article misses the mark. The withdrawal of Fable is unlikely to have pushed many people into the arms of open source just yet. Claude Opus 4.8 remains far ahead of current open-source models and is a favorite with developers. There is a growing undercurrent of teams wanting to hand some workloads off so they can be cheaper, and ideally local, but I don’t think it has much to do with the Fable restrictions. This incident certainly reminded everyone that API access can be revoked, but the long-term issue is much more likely to be driven by cost and privacy concerns than guarantees for access because models are still quite easy to switch out. A case in point is that open-source models like GLM-5.2 are now getting very good at coding.  
• Ben Thompson and Michael Morton on eCommerce in the Age of AI. Great up-to-date interview that covers current trends in web commerce as AI influence shifts. The big takeaways are: 1) AI is still functioning more as a referrer than an aggregator (which will be a relief to existing marketplace players), and 2) Shopify is in a potentially stronger position than most people realize. However, I disagree with their assessment that AI won't solve the "whole basket" problem soon. I think this is relatively trivial to solve if you can build long-running persistent order lists and allow agents to disaggregate baskets invisibly in the background. The conversation goes into the dynamics of autonomous vehicle ride-sharing.

Wishing you a great weekend.