Thanks for sharing your experience, Daniel! I’ve been meaning to try out Claude Code for a while now and this pushed me over the edge to jump in and get going with it.
Nice straightforward tips, and demystified it a little for me. 👍
great post! I’ve also found that when working with UI, using Playwright mcp to capture screenshots for pixel-perfect checks or to grab console logs works really well
One thing that trips people up with pasting screenshots: long or dense ones get squeezed down hard before the model ever sees them. A tall page can lose most of its detail once it's resized to fit the pixel limit, so precision disappears right when you need it most, like reading small text or picking the exact element you meant. Full page capture plus a structured handoff instead of a flat image fixes this. Wrote up the actual numbers here if useful: https://slimsnap.ai/blog/screenshot-downscaling
The point about Claude Code as more than a coding assistant resonates. I use it to orchestrate an entire personal automation system - scheduling, memory, multi-channel integration. The harness around Claude Code is where the magic happens. https://thoughts.jock.pl/p/wiz-personal-ai-agent-claude-code-2026
The real strength of agent mode is when agent can check the output (logs, screenshots) itself.
E.g. I added playwright MCP server so that agent itself can load the site and take screenshot or investigate the DOM tree to check it it achieved the result.
I’ve seen several of these posts now about devs moving exclusively to Claude from Cursor… but my experience is that Claude is a massive token hog, it eats up a ton of money for small outputs compared to other tools. Do you have a workaround for this or a different experience?
Thanks for sharing your experience, Daniel! I’ve been meaning to try out Claude Code for a while now and this pushed me over the edge to jump in and get going with it.
Nice straightforward tips, and demystified it a little for me. 👍
Thanks brother, feel free to DM me if you have any question about it!
great post! I’ve also found that when working with UI, using Playwright mcp to capture screenshots for pixel-perfect checks or to grab console logs works really well
Thanks nice post
Also currently I have using qwen cli free version I did used claude code in past
One thing that trips people up with pasting screenshots: long or dense ones get squeezed down hard before the model ever sees them. A tall page can lose most of its detail once it's resized to fit the pixel limit, so precision disappears right when you need it most, like reading small text or picking the exact element you meant. Full page capture plus a structured handoff instead of a flat image fixes this. Wrote up the actual numbers here if useful: https://slimsnap.ai/blog/screenshot-downscaling
I’ve been trying out vibe coding tools
https://iruafeimi.substack.com/p/my-thoughts-on-vibe-coding?r=57lqd9&utm_medium=ios
The point about Claude Code as more than a coding assistant resonates. I use it to orchestrate an entire personal automation system - scheduling, memory, multi-channel integration. The harness around Claude Code is where the magic happens. https://thoughts.jock.pl/p/wiz-personal-ai-agent-claude-code-2026
Cursor was the first agentic ai that blew my mind but then Claude Code did that on another level and haven’t turned back since
3 weeks already passed from this note… and this might change already, an other one is the best 😂
> Give Claude screenshots
The real strength of agent mode is when agent can check the output (logs, screenshots) itself.
E.g. I added playwright MCP server so that agent itself can load the site and take screenshot or investigate the DOM tree to check it it achieved the result.
Cursor uses sonnet and gpt models at it's end
Amazing to have custom windsurf.md, claude.md etc
I’ve seen several of these posts now about devs moving exclusively to Claude from Cursor… but my experience is that Claude is a massive token hog, it eats up a ton of money for small outputs compared to other tools. Do you have a workaround for this or a different experience?
really helpful thanks for sharing this Daniel.
Um, interesting, I didnt know you can put Claude markdown in each folder...🤔