Hey I’m Ben. I build stuff with agents, even though I’m not technical. Here’s all the stuff I’m reading and tinkering with. If you want to start building or level up your ‘vibe-coding’ skills, join our community.
Hey folks,
Anthropic has revised its “responsible scaling policy”, which is arguably a bit more flexible now—potentially allowing them to keep building new models beyond the “accepted safety limits”. This is separate from another similar news: The US Department of Defence wants more relaxed access to Claude, but Anthopic is denying it based on their two principles for military use → 1) no autonomous weapons and 2) no mass surveillance of Americans. If they stay firm on this by Friday, they might lose a $200M government contract.
In the meantime:
You can now schedule tasks in Claude Cowork. Plus, enterprise users can now create plugins for Cowork and share within their org to customise it.
Claude Code got remote control - Any Claude Code session started in the terminal can be later accessed remotely through the Claude mobile or web app. It keeps working on the original machine (the one where it was initiated).
Cursor Agents now use a computer to test their work and return a video demo of their output. Also, Cursor acquired Autotab & Anthropic acquired Vercept — both acquisitions mention a goal to make better computer-using agents.
Perplexity also released Perplexity Computer - A general agent with tools for research, design, code & more (see examples), and Google previewed Gemini using apps on an Android device to order food autonomously at Samsung’s Galaxy S26 launch.
Many “agentic” tools are moving from using the computer via CLI to letting the agents use the computer like a human. Clearly the next big thing…
Trust-First AI. Built Into Your Browser. Norton Neo is the AI-native browser built from the ground up for privacy, security, and speed. It adapts to you, organizes your tabs, and works without constant prompting. Experience the future of browsing.*
🌐 What I’m consuming
Opus 3 has a blog - Claude’s Corner.
Building Bloomery - an agent skill that coaches you through building an agentic loop yourself.
Teaching Codex editing in Premiere with a plugin that it wrote.
How we hire engineers when AI writes our code?
Mitchell Hashimoto on how AI is changing his day-to-day work.
I vibe-coded my dream macOS presentation app.
Two beliefs about coding agents. 1) Most talented developers do not appreciate the impact of the intuitive knowledge they bring to their coding agent, and 2) Most work people are sharing are incredible personal tools, but they are not capital-P products.
Airia Hackathon: Race Beyond the Track with Atlassian Williams F1 Team. Join Airia’s Virtual AI Agent Hackathon Feb 23–Mar 1 ahead of the Australian Race weekend. Build autonomous agents that elevate the fan experience. Top projects will be highlighted before the first race weekend of 2026.*
⚙️ Tools and demos
Flint - Build a unique page for every ad, keyword, and customer without bothering engineering.
Custom Agents in Notion - Autonomous AI bots that can triage items, answer questions, draft reports, etc. Teams like access with on-schedule or 24/7 operation.
Quiver AI - Foundational tools for generating, editing and animating vector graphics (SVGs). I tried it on a couple of images with curved lines and compared it with Gemini’s output. Quiver is a lot better.
Paper - A design canvas for Cursor, Claude Code, Codex. Any agent can read and write HTML to Paper.
Boards by Variant AI - Curate inspiration and turn it into design instantly.
Hermes - An intelligent markdown editor with AI that will never write for you.
Typeless - Clear voice transcription without fillers on all your devices.
Raindrop AI’s new “self-diagnostics” feature lets agents proactively self-report issues they encounter.
🥣 Dev Dish
GPT-5.3-Codex is now available in the Responses API.
Apple released a Python SDK for Mac’s on-device LLM.
GitAgent - A git-native, framework-agnostic, open standard for defining AI agents.
Better Hub - Reimagining GitHub as a better place to collaborate on code for humans and agents.
Devin 2.2 - Let your agent test its work with computer use, self-verify, and auto-fix any mistakes.
Hermes Agent - Open source CLI agent that’s somewhat between Claude Code and OpenClaw.
Browser Use Agent API - Complete complex tasks on the browser with a single API call.
🍦 Afters
Another AI chip company, MatX, has raised $500M series B to build LLM specific chips for frontier labs.
Amazon plans to invest $50B in OpenAI, but $35B of it is only if OpenAI goes public or builds AGI.
Google Labs acquired Producer AI, a music generation tool that launched in Aug 2025. It now uses Google’s Lyria 3 for music generation. Labs also moved its experiments, Whisk and ImageFX, into its video creation product “Flow”.
Enjoy this newsletter? Forward it to a friend.
That’s it for today. Feel free to comment and share your thoughts. 👋
Read about me and Ben’s Bites
📷 thumbnail by @keshavatearth
* sponsors who make this newsletter possible :)
Wanna partner with us for March? Last few slots available
The shift from CLI agents to computer-using agents feels like the real inflection point here - validating work at the UI level is where most agent products sink or swim. On an unrelated note, a tool I've found genuinely useful is https://mangatranslator.me - it translates manga pages while keeping the original layout intact.
Respect to Anthropic for holding firm on no autonomous weapons and no mass surveillance, even with $200M at stake - that consistency builds real trust in AI labs. The computer-use race is moving fast too. Also enjoyed the image tools roundup; on that note, I've been merging shots with https://aiimagecombiner.app, a simple two-image combiner that works well.