What the people actually building AI said today. One page — a 2-min read.
Today’s strongest signal is about making agents useful in real work: they need secure access, persistent workspaces, and ways to check their output. Builders also shared where model choice and product design still get in the way.
Legora tests legal agents without seeing clients’ confidential work
Legora says its agent can do roughly 60% of current legal work well, but the team cannot inspect most customer data to see where it fails. “We can't look at any traces, basically,” Jacob Lauridsen says. Its engineers and lawyers build task-specific evaluations, while Tabular Review checks the first 10 rows before scaling a contract review and links results to source documents.
Rauch says agents need a workspace that follows them
Guillermo Rauch argues that stateless sandboxes miss what makes agents useful: access to their identity, skills, repositories, and memories. He says Vercel Sandbox is operating in 19 global regions while meeting latency and throughput targets amid double-digit weekly customer demand growth.
Tan reports a fivefold jump in issue-fixing throughput
Garry Tan says Capy Captains increased his issue-fixing throughput by about 5X overnight. He credits the ability to coordinate work across captains and subthreads.
Yang says Gemini’s edits beat ChatGPT’s, but the workflow stalled
Peter Yang found Gemini’s suggested copy edits useful and ChatGPT’s mostly poor on a post he was writing. He says Antigravity’s browser use did not work for the task, leaving him to paste Gemini’s edits into ChatGPT to apply them.
Sottiaux says the dot will choose models by task
Thibault Sottiaux says “your dot” will select the best model for a task and draw on different models for different parts of the work. He gives no further detail about how that selection will work.
Shipper credits Argon’s quality and Antigravity’s simpler design
Dan Shipper says Argon’s appeal comes partly from its quality and partly from Antigravity lacking newer features and complexity. He points readers to Every’s Argon review for more context.
Source data comes from the open-source project follow-builders by zarazhangrui, released under the MIT license. Summaries are generated by an LLM from that project's public feeds, and the summarization prompts are adapted from it. Every item above links to its original source.
Summaries generated automatically. Read the original before relying on any claim.
