Claude subscribers get monthly credits to build agents

Welcome back. Every developer knows the sting of paying $200 for Anthropic's Max plan only to shell out even more for API credits. Anthropic just resolved that headache for you. Meanwhile, a new AI hardware startup rolled out a wearable smart ring that lets you prompt your agents directly from your finger. 

Also: One hacker and agents may be behind South Korea’s bank hack, a dev used AI to discover a possible planet, and a pro trick to make your Grok Bots smarter.

Claude subscribers get monthly credits to build agents: The AI lab's Max plans now include $100 or $200 in monthly API credits, while Team plans pool up to $500. They cover the Claude API, Agent SDK, and Managed Agents. Anthropic engineer Thariq Shihipar used his to build a Chrome new tab page that Claude rebuilds daily from his browsing history, then open-sourced it. Anthropic is also tightening its usage rules next month, banning sustained cruelty toward Claude but exempting model testing.

This new startup wants you to prompt coding agents from a ring: AI startup Natura just unveiled Interface, a $99 smart ring that lets you send voice requests to Claude Code, Codex, and other agents. Just press, speak, and release to kick off a task without reaching for your laptop. It can also record meetings and control your computer. Preorders open next month, with a $9 monthly subscription kicking in after an initial period. Shipping is slated for December or January. Natura is also unusually upfront that specs may change before release. Join the waitlist.

Paul Graham calls Amazon's agent ban a startup opportunity: Last month, Amazon asked Meta to pull its store off Muse, Meta's personal agent. When Meta refused, Amazon flat-out blocked the agent. The Y Combinator cofounder thinks that move gave competitors their first real opening since Amazon launched.  His logic is simple: platforms don't ban agents people aren't using, which means there's untapped demand up for grabs. Some devs are already debating whether agent-friendly checkout APIs are the next big play, while others are completely against it.

Datadog’s latest AI guide explains how to move from experimental AI “bets” to measurable production systems that drive business impact.

Written for CTOs, CIOs, and engineering leaders

No more gut feelings; Prove your AI projects with real production impact. 

Read the guide here.

Spending got a little too easy. Django co-creator Simon Willison is making the case for hard budget caps on every pay-per-use service and API, switched on by default. Once you hit your monthly limit, the service simply shuts off. His argument is that a simple warning email just does not cut it anymore. 

Your agents don't stop when you go to bed. Coding agents, and the friendlier personal agents built around them, make it easier to ship code that calls paid APIs, hosts apps, and racks up storage and compute charges. But those bills can keep climbing while you sleep. An alert lands at midnight, and by the time you read it, a rogue service has burned through hundreds or thousands more dollars.

Cloud providers are starting to catch up. That’s not to say there haven’t been efforts to stop this. Google Cloud launched Spend Caps in public preview in July, letting teams limit monthly spending on a specific service within a project. AWS followed in September with project-level limits that pause a project for the rest of the month. There is a catch, though. AWS is still rolling the feature out to a limited group of customers, and both caps require you to opt in. 

An error beats a $10,000 surprise. Nobody wants a production app to go down because it hit a spending limit. Still, Willison argues most teams would take downtime over a surprise $10,000 bill. He wants hard caps turned on by default, with a clear opt-out for anyone who needs to lift them. He also thinks agents should actively recommend providers that support those caps. Until they do, make sure yours flags any uncapped paid service before you deploy. 

Try it: Google's Spend Caps guide shows how to cap the Gemini API or Cloud Run for one project. You should set the cap below your real limit, since enforcement isn't instant.

Granola, the AI notepad for engineers in back-to-back meetings, now takes notes in 32 languages.

Run a standup in Spanish, a design review in Japanese and a customer call in English, and Granola turns each one into clear summaries and action items.

Use code THECODE for your first month on us.

How to build a full agentic development stack: A dev walks through the harnesses, models, dev tools, and workflows he actually uses to build software with agents. You’ll learn how to pick between Claude Code, Codex, Cursor, and other tools, then tie them together into a software-factory-style workflow instead of relying on one coding agent for everything.

REA (33.1k ⭐): Give your coding agent reverse-engineering superpowers. REA lets Claude Code, Codex, Cursor, Gemini, and others inspect apps, websites, binaries, Electron projects, APKs, firmware, and more, then return the evidence behind how something works so your agent can explain it or rebuild the feature.

How to build reliable agent automations (by Anthropic): This guide shows how to build background agents that run on a schedule without quietly missing things or repeating themselves. You’ll learn how to track where each source left off, handle failed reads, keep memory clean, verify actions before recording them, and cap what each run can spend.

Running Opus burns through usage limits. Running Haiku alone means dumber answers on hard problems. A senior dev shared a settings tweak that gets both: Haiku does the work, Opus checks it.

A senior engineer claims AI-written tests are empirically useless after his agent solved more tasks without them. Here's what he says to tell your agents instead.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 350K+ engineers and 150K+ followers on socials. Get in touch.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team