9 items9 builders

AI Builders Digest

What the people actually building AI said today. One page — a 4-min read.

Vercel's AI Gateway numbers show Anthropic's share of spend sliding from 69% to 40% in two months while OpenAI more than doubled, the clearest read yet on how fast frontier-model demand actually moves. Underneath the model layer, VAST Data's CEO makes the case that the storage-and-software middle tier is where the value accrues, and that agents will soon fine-tune their own models on enterprise data. Anthropic and Replit both pushed product at the low end: 43 small-business workflows, and Muse building apps on Replit.

X

Guillermo Rauch

Vercel CEO

Vercel gateway data: Anthropic's spend share fell 69% to 40% in two months

Two months of Vercel AI Gateway traffic shows Anthropic still leading in spend but dropping from 69% to 40%, with OpenAI climbing from 10% to 24% and now leading on raw token count. Kimi K3 and DeepSeek absorbed roughly half of what Anthropic lost, which makes this as much an open-weights story as an OpenAI one. Opus 5.5 reached up to 10% of spend within two days of launch, and OpenAI accounts for 62% of image generations.

  • #products
  • #open-source
Podcast

The MAD Podcast with Matt Turck

VAST Data CEO: enterprises will own the weights, not rent the intelligence

Renen Hallak argues the reason to build your own AI factory is not cost, it is that your IP ends up distilled into weights, and handing your processes to a model provider means handing over the secret sauce. His framing: "we infer during the day, we fine tune at night", with agents accumulating experience the way a person does and every organization eventually owning the models its agents produced. VAST's new DataEnclave runs inference inside the enterprise with the model builder's weights encrypted end to end through NVIDIA's encrypted memory, so neither side sees the other's asset. On demand, one AI cloud told VAST it needed 500 petabytes over three years, then came back last week asking for two extra exabytes on top: "sometimes it scares me."

  • #agents
  • #hardware
Blog

Claude Blog

Claude for Small Business adds 43 workflows and 27 integrations

Anthropic expanded Claude for Small Business from back-office cleanup into revenue work, shipping 43 workflows and 27 new integrations covering Shopify, Salesforce, Stripe, Square, Xero, Gusto, TikTok, Zoom, Atlassian, and Zapier. The product has been installed more than 900,000 times since launching in May, and the roadmap came straight from over 1,000 owners across 10 cities, about a third of whom asked specifically for lead generation, inbound replies, and proposal writing. One owner reports making $20,000 in six weeks on Claude-written proposals; another runs a 6am CRM sweep that emails the day's priorities. Free workshops return this fall in 10 US cities, with 150-plus trained partner organizations running over 750 more.

  • #products
X

Thariq

Claude Code plan mode is becoming a built-in mod you can override

Feedback split cleanly: plenty of users plan their own work and want plan mode gone, while others like having a mode where Claude only thinks and brainstorms. The resolution is to turn plan mode into a built-in mod and let mods add new modes or override shift+tab entirely, so you can customize the plan prompt, publish your own modes, or rebind the key to something else.

  • #products
  • #agents
X

Peter Steinberger

Tell an agent to cut 20% of tests, not to "clean up"

Vague cleanup instructions make agents stop far too early. The fix is an ambitious, measurable goal: "remove 20% of the least useful tests while maintaining code coverage within 2%" gives the agent both a target and a guardrail. He also ran Daybreak across open-source dependencies and turned up 8 more long-standing leaks.

  • #agents
X

Swyx

"Scaling without Slop" worked: 100k YouTube subs in 1.2 months vs 3 years

The content strategy swyx called at the start of the year is now compounding. The first 100k YouTube subscribers took three years; the next 100k took 1.2 months, with similar traction on AEO/SEO and subscribers. He is using the milestone to announce a next phase for Latent Space, AINews, and the rest of the portfolio.

  • #products
X

Amjad Masad

Replit CEO

Muse can now build apps on Replit

Replit has wired Muse into app creation, so the model can now produce working apps directly on the platform.

  • #products
  • #agents
X

Josh Woodward

Google VP

Google Labs' Dreambeans rations a fixed number of "beans" each morning

Dreambeans is a newer Google Labs experiment with a growing cult following, built on deliberate scarcity: a fixed number of beans brew every morning and each one points you at something real happening offline with people you care about. It is a bet that constraint, not infinite feed, is what makes a recommendation product worth opening.

  • #products
X

Nikunj Kothari

The bet on trades software: every owner gets a bespoke app

The thesis he is investing behind is that every small business owner, specifically in the trades, will end up with bespoke software rather than a seat in someone else's SaaS. The differentiation lives in the last mile, in the experience an owner can give their own customers and employees, which is exactly the part off-the-shelf tools cannot cover.

  • #products
  • #funding

Get this in your inbox

One email a day. Unsubscribe in one click.

Where this comes from

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.