Welcome back. AI agents can reason, plan, and code, but getting them to work reliably with your local desktop apps is challenging. Vercel just changed that with a new release — and all it takes is one line of code to set up.

Also: How one lawyer used Claude skills to outperform 100+ law firms, what Altman revealed in his Pentagon AMA, and the agent mistakes Anthropic doesn't want you to repeat.

Vercel gives AI agents control of desktop apps: The developer platform just dropped a new Electron skill for its agent-browser tool, giving AI agents the ability to click, type, and pull data from desktop apps like Discord, Figma, Notion, Spotify, and VS Code. Developers can set it up with a single command (npx skills add vercel-labs/agent-browser --skill electron) through the Chrome DevTools Protocol, essentially turning any Electron app into an API. It's an easy way to automate testing, debugging, and repetitive workflows.

White House blacklists Anthropic over military safety dispute: After Anthropic refused to lift Claude's safety guardrails for military use, the Trump administration labeled the company a "supply chain risk" and ordered all federal agencies to stop using the model. The Pentagon is now severing its $200 million contract and requiring contractors to certify that they aren't using Claude. In response, the department has already pivoted to OpenAI with the offer, even though the ChatGPT maker maintains similar safety standards.

Cursor's code review agent now fixes bugs automatically: The AI code editor just upgraded its code review agent, Bugbot, which can now automatically fix bugs found in PRs. The new Autofix feature launches cloud agents in virtual machines to test code and suggest patches. More than 35% of these proposed fixes are getting merged, and the tool's bug resolution rate has climbed from 52% to 76%. It's now live for all users.

Delve is the AI-native compliance platform that actually does the work for you, auto-collecting evidence from AWS, GitHub, and your stack so you’re not chasing screenshots or babysitting integrations. Use AI security questionnaires and an AI copilot to make compliance less dreadful.

The proof is in the pudding:

Free migration. Zero disruption. No starting over.

Book a demo, trigger your migration, and get $2,000 when you’re onboarded.

New workflows are arriving faster than teams can keep up. Cursor CEO Michael Truell recently laid out the three stages of AI coding: Tab autocomplete, synchronous agents, and now autonomous cloud agents that work on their own VMs and submit fully tested pull requests. Tab autocomplete stayed the standard for almost two years, but synchronous agents might not even last for one.

The new New Thing. In March 2025, Cursor had two and a half times more Tab users than agent users. That ratio has completely flipped since then, with agents’ usage growing 15x in just a year. Today, 35% of the pull requests Cursor merges internally come from autonomous cloud agents that produce artifacts, resolve merge conflicts, and ship everything on their own.

Mo workflows, mo problems. If you’re feeling overwhelmed with all these new tips and tricks and endless new ways of doing things, you’re not alone. A simple solution: Andrej Karpathy recommends an 80/20 split, where you spend most of your time in whatever workflow is working for you right now, while setting aside a consistent chunk to explore the next phase, even if it feels a bit unpolished. The goal is to build teams that see their current process as a work in progress, not the end result.

Code migrations are inherently tedious. Even though the same patterns repeat across every file, you're usually stuck handling them manually, which is why those tickets often sit in the backlog for months.

The latest Claude Code update addresses this with two new commands, which creator Boris Cherny shared last week.

"/batch" condenses a migration into a single line. It launches dozens of parallel agents (each in its own git worktree) to run tests and open a PR once finished.

“/simplify” reviews your recent changes using three parallel agents focused on code reuse, quality, and efficiency. It then aggregates the findings and auto-fixes any issues it detects.

Both commands work right out of the box. You can even chain them together, letting /batch handle the migration while /simplify cleans up the output in one go.

How to build coding agents (by an Amazon engineer): A 10 step guide for building AI agents that can independently manage tickets, write code, and handle reviews. This tutorial shows devs how to move past manual prompting by using MCP servers, reusable skills, and SOPs to create specialized agents that work around the clock.

Awesome OpenClaw use-cases: A collection of 34 real world OpenClaw use cases, covering everything from automated code reviews to Jira workflow management. It's a great resource for devs looking to get some actual value out of AI coding agents.

Codified Context: AI coding tools struggle with large codebases because they often lose context between sessions. By structuring project knowledge into a three-part memory system (always-loaded rules, specialist agents, and on-demand docs) you can help agents retain context and write more reliable code.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 200K+ engineers and 100K+ 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