Thoughts.

Nano Banana 2 just dropped and my MCP server already supports it.

Google released Nano Banana 2 today: almost Nano Banana Pro capabilities at Flash speed. Subject consistency across up to 5 characters, text rendering that actually works, resolutions up to 4K, and it's grounded in real-time web search so it can render specific subjects accurately.

After testing it I am finding that the quality doesn't quite match Nano Banana 3 Pro. You can tell. But that's not really the point. Because it's much quicker and cheaper, you can generate more variations in the same amount of time, which makes it fantastic for brainstorming, exploring directions, and rapid iteration. When you need the final polished version, go back to Pro. But for getting to that final version faster? Nano Banana 2 is the faster tool.

I updated my open-source Image Generation MCP Server to support it. If you're using Claude Code, Cursor, or any MCP-compatible IDE, you can generate Nano Banana 2 images at any aspect ratio directly from your editor, no square-only limitation, no leaving your workflow. I also added image editing capabilities, so you can pass in existing images to edit or use as reference, similar to the web grounding feature.

Claude Code just launched custom slash commands. Drop Markdown files into .claude/commands/ and you get repeatable, templated prompt workflows. Type /project:fix-github-issue 123 in the CLI and it injects your $ARGUMENTS directly into the prompt, automating issue triage, code reviews, refactoring, and more.

Inspired by this, I put together a GitHub repo of curated slash-command prompts. Generate tests, audit security, format docs, streamline Git flows. Clone the repo into .claude/commands/ and start typing /project:… in Claude Code.

What workflows have you automated with Claude Code?

Claude Code just fixed my overheating MacBook. After a bit of prompting, it ran a couple of commands and found out that a Chrome extension was bugged and keeping my laptop awake, causing it to overheat.

It feels like the future is already at our fingertips. You just have to know how to use it.

If you haven't tried the Codex CLI from OpenAI yet, you definitely should. I'm finding it's meaningfully better than Claude Code with Opus 4.1 at certain tasks, especially its ability to make targeted fixes and edits rather than producing a flood of changes.

Make sure you select GPT-5-high in the model selector for maximum reasoning.

Seeing as many of the AI providers and AI coding tools are currently down due to an outage, now is the perfect time to explore running large language models locally. LM Studio and Ollama are what I'd recommend to get started.

I even built an open source AI coding assistant that runs totally offline: no cloud, no outages, no data leaving your machine.

Testing out the new ChatGPT 4o image generation for creating marketing graphics. So close to being good but it keeps cutting off images or making typos. Still an insane progression from the previous version.

Couldn't fit all 7 steps into the image, but still managed to get the point across. Guess you gotta go to the website to check out the 7th step.