An open-source harness for agents just landed.

Bankr moved $900,000 through Robinhood in its first week.

This Week

Harrison Chase released deepagents, a model-agnostic toolkit that lets developers wire, run, and monitor AI agents with a few lines of code. The repo includes dynamic subagents for programmatic workflows and plugs into LangSmith for tracing.

Previous agent demos broke when you swapped models; deepagents separates the agent logic from the underlying model the way Docker separates apps from servers. Teams can now switch between GPT-4o, Anthropic Claude, or open models without rewriting code.

→ Look for the first public production case study from an early deepagents adopter before the month ends.

By the Numbers

22.8M

downloads of Vercel Labs AI tools

1T+

tokens Vercel AI Gateway recovers monthly

100K+

Bankr transactions to date

$50M+

$BNKR market cap

335K+

OpenClaw GitHub stars in 60 days

Signals

Guillermo Rauch: Vercel ships Eve eval for agent testing out of the box.

Harrison Chase: launches deepagents, an open source model agnostic agent harness.

0xDeployer: Bankr reports $900,000 volume on Robinhood launch.

0xDeployer: Bankr launches paid API for lottery game with $1,000 prize.

0xDeployer: Bankr supports limit orders and trailing stop loss on Base.

Austin Griffith: releases one dollar audit for smart contracts.

Peter Steinberger: built nameplate for clearer screen sharing.

Builder to Watch

0xDeployer@0xDeployer

If you're not following 0xDeployer yet, Bankr stacked three trading features and a lottery API in seven days. Next up: watch $BNKR liquidity as Robinhood volume climbs.

Week in Review

deepagents gave devs a swap-friendly agent harness, and Vercel answered with Eve eval for instant grading. Bankr piled on new trading toys and still rang up $900,000 in Robinhood volume. A $1 smart-contract audit and a screen-share nameplate rounded out the week.

→ Watch for the first public contracts audited through Austin Griffith's $1 audit tool.

Shipped This Week