Top pick: "The Amazon Trap" by @mvernal, recommended by Sarah Guo and Scott Belsky

What it argues: Vernal's essay is billed as "Why Amazon is blocking Instinct & Muse, and one potential path forward" @mvernal. He says horizontal personal agents will win because they have "the totality of their context," including email, calendar and files. That leaves Amazon with two choices. It can give up a core strategic position and up to $80B a year in ad revenue, or it can antagonize its customers @mvernal. His proposal is a "Prime+" tier at roughly $299 a year, with agentic purchasing available only to its members @mvernal. Disclosure: his firm was Instinct's initial investor, sits on its board and seeded the team now working on Muse @mvernal.

Why they shared it: Guo called it "some strategic wisdom from our partner" on "the choice of whether to resist the future or rapidly adapt" @saranormous. Belsky quoted the line "Companies often anchor to their priors." He added that "you can't fight customer preferences for horizontal agents, you must innovate to meet them," and said he'd like to hear what Andy Jassy thinks @scottbelsky.

Bill Gurley: "Machines of Loving Grace" by Dario Amodei

This follows Gurley's push for the NYT piece on Anthropic, covered in the last brief. A reply argued that Anthropic's approach differs from other labs' and that, under it, "the machine will inevitably control humans" @Benk878. Gurley pointed to Amodei's essay and said the view is "directly stated" there. He added that the company makes job candidates read the essay and "even hands out to customers" @bgurley. He doesn't say whether he agrees with it. His point is that the essay is the place to understand how Anthropic thinks.

Paul Graham: Christian Szegedy's personal essay on AI and mathematics

Szegedy says he wrote the essay after reading recent essays by Terry Tao, Kevin Buzzard and Timothy Gowers @ChrSzegedy. Graham shared it with this quote: "The skeptics who say AI can't conjecture, can't theorize, and can't explain are describing last spring's models. Within months, they will be describing nobody" @paulg.

Amjad Masad: a16z podcast with OpenRouter's Alex Atallah

Masad is a guest on this episode and co-founded Replit, so it is partly self-promotion. He said the conversation goes "really deep and somewhat technical" and called Atallah "brilliant" @amasad. It is Atallah's first podcast since Stripe acquired OpenRouter @a16z. Topics include:

• how the Stripe deal came about
• why "payments and inference are going to blend together"
• why enterprises are spreading their usage across labs and open-weight models @a16z

The two disagree on personal agents. Masad runs one agent across his whole company. Atallah argues that "10 specialized chiefs of staff beats one superagent" @a16z. YouTube@a16z.

Shorter mentions

• The Best of Om Malik by Elijah Potter (@theleastsirius), recommended by Matt Mullenweg. Potter put it together after discovering Malik's archives. Mullenweg said it makes him "very happy that a new generation is revisiting Om's writing." He added that Malik's work "really shaped a lot of my thinking" @photomatt. This follows his praise last week for another tribute to Malik.
• Social Capital's post on the economics of aging, shared by Chamath Palihapitiya. Social Capital is Chamath's own firm. He framed the post as "The case for giving everyone GLPs…" @chamath. The post estimates that slowing aging by one year would add about $400B a year to U.S. GDP, and that slowing it by five years would add $2T. It sets this against Social Security and Medicare spending of roughly $2.6T in 2024 @socialcapital.