AI Builders Digest

Claude Fable 5.2 reaches paid plans; Mythos access opens by application

Claude Fable 5.2 is available today on all paid plans. Organizations working on beneficial cybersecurity or biomedical research can apply for access to Claude Mythos 5.2. Anthropic also says Fable 5.2 found the exact limit of a technique used to rule out candidate theories of quantum gravity.

• #products
• #research

Fable 5.2 enables ambitious builds, but its cost changes the calculus

Dan Shipper's team used Fable 5.2 to build a custom video editor, turn a child's toys into a racing game, and make a documentary from research. He recommends giving it a large job with a clear definition of done and time to test its work. The catch: it is slow and token-heavy, so he would choose another model for routine writing or quick coding exchanges.

• #products
• #coding

Handshake's benchmarks expose costly gaps in AI agents

Handshake AI says its life sciences benchmark has even frontier models at 50% accuracy or worse, while agents score below 20% on its simulated finance tasks. A banking agent produced a convincing spreadsheet but mishandled one tax rule, corrupting the valuation without flagging uncertainty. As one banker put it, “If it's 95% right, it's 100% wrong.” Handshake argues that realistic tasks, expert grading, and tests for harmful behavior are essential before these agents can be trusted.

• #evals
• #agents
• #safety

Vercel is adding event subscriptions to its MCP

Guillermo Rauch says MCP Events fills a gap in the protocol: an agent can subscribe to a later event, such as a successful deployment, instead of repeatedly checking for it. Vercel is adding support to its own MCP. Rauch also argues that newer models can coordinate more parallel work than one computer can handle, making cloud sandboxes increasingly useful.

• #agents
• #open-source
• #products

Dots can now control multiple machines

Dots can now control many machines, extending beyond the cloud computer it provides by default. That gives the agent a broader set of computers to work across.

• #agents
• #products

Aaron Levie expects coding-style AI gains across knowledge work

Aaron Levie predicts that advances in coding, computer use, and reasoning will bring autonomous collaborators to fields including animation, manufacturing, architecture, and product design. He expects those industries to see output gains similar to software engineering's, with demand expanding as production becomes easier.

• #agents
• #industry

AI could make product managers deeper across the whole lifecycle

Madhu Guru argues that AI should help product managers work more deeply at every stage, from user research and prototypes to codebase exploration, analytics, and go-to-market decisions. Her suggested exercise is to map where PMs spend their time today and identify the handoffs where information gets lost.

• #products
• #work

Longtime subscribers get more Fable 5.2 resets

Anthropic is tying Fable 5.2 resets to subscription tenure. Pro users get one reset; Max, Ultra, and Team Premium users can get up to three, depending on how long they have subscribed.

• #products