Hello, I’m Ben. I enjoy creating tools with agents, despite not having a technical background. Here’s a collection of insights and projects I’m exploring. If you’re looking to start building or enhance your ‘vibe-coding’ skills, join our community.
Greetings, everyone!
Claude Sonnet 4.6 has been released. This latest version outperforms Opus 4.5 on most benchmarks and even exceeds Opus 4.6 in two categories: office tasks and financial analysis. Additionally, it’s quite proficient in browser and computer-based tasks. If you’re using simple agents, consider switching to Sonnet 4.6 to maximize your capabilities.
Sonnet 4.6 is also the new default model for free Claude users. Until now, I recommended ChatGPT to those outside the AI community because Claude’s free tier didn’t offer much in terms of computational power or features. However, this update introduces numerous enhancements, such as file creation and connectors, to the free tier. Claude has improved its web search functionalities as well, reducing the clutter in the context window.
And as is often the case with AI, there’s a bit of *drama* involved…
Firstly, some background: Anthropic and OpenAI both provide heavily subsidized plans at $200/month (around 15 times cheaper than using APIs). However, they are approaching third-party developer access quite differently. Earlier this year, Anthropic assured developers that using the Agent SDK (previously known as Claude Code SDK) with a Claude subscription was acceptable, but they recently revised their documentation to indicate otherwise. Since then, they have been vague about whether open-source apps can allow users to utilize their own subscriptions. Theo offers a detailed analysis here. In contrast, OpenAI has openly supported third-party use of Codex through ChatGPT OAuth. The primary concern is that Anthropic’s unclear policy changes and insular practices are creating significant uncertainty for developers looking to build on their platform.
Gemini can now produce music. Google’s latest music generation model, Lyria 3, is integrated into Gemini and can create songs with lyrics based on your prompts, images, or even videos. It generates 30-second clips accompanied by art created by Nano Banana.
I experimented with it for a short while.
a) it’s speedy,
b) the output can be a bit cringeworthy. However, I’ve noticed a pattern in AI-generated music on YouTube, and the results are somewhat reminiscent of popular styles from about six months ago, when the quality started to improve. Genres like study or relaxation music are now thriving on YT.Two samples from my experiments demonstrate how Gemini handles lyrics, ambiance, and copyright issues: a techno chant and a techbro lullaby
— Keshav
Claude Code has been integrated into Figma – The Figma MCP now allows you to design using Claude code before sending it to Figma, where you can refine it using familiar tools.
Attio is an AI CRM designed for modern go-to-market teams. Sync your email and calendar to effortlessly create an enriched CRM that offers extensive context. After that, simply ask questions about meeting prep, call insights, or any business-related inquiries. Join the ranks of fast-growing teams like Granola, Flatfile, and Modal. Start for free today.*
-
Speechmatics – Speech-to-text solution for voice agents. BB readers can claim $200 in free credits.*
-
Traces – A CLI and web tool for sharing and discovering sessions with multiple coding agents.
-
Intent by Augment Code – Manage your projects and orchestrate agents all in one place.
-
Aperture by TailScale – A gateway for LLMs to centralize model access and monitor team usage without managing individual API keys.
-
Lemon – Engage with your computer to have it perform tasks for you. It goes beyond simple transcription to include sending emails, calendar management, research across tabs, and more.
-
Monologue, the transcription app I use on Mac, is now available on iOS as well. I’ve tested it extensively, and it’s fantastic!
-
Cursor Marketplace – Discover and install plugins that bundle skills, MCPs, subagents, hooks, and more, covering the full development lifecycle.
-
Wiretext and Mockdown – Both platforms enable you to create quick wireframes and share them as markdown or ASCII format with your coding agents. (wiretext demo — mockdown demo)
-
Polsia autonomously builds clones (essentially landing pages with a “get early access” button) for popular tools, having already created over 500 of these to date.
-
EVMbench – a new evaluation tool from OpenAI designed to test whether models can exploit and patch smart contracts on blockchains. Most models can identify several vulnerabilities, patch only a few, but are capable of exploitation.
Thanks for reading this newsletter! If you enjoyed it, consider forwarding it to a friend.
That’s all for today! Feel free to share your thoughts in the comments. 👋
* sponsors who made this newsletter possible 🙂
Interested in partnering with us for Q1?
