Designing and building a hook-based system that automates Claude Code's repetitive permission prompts. Preset-driven configs, learning from manual approvals, and safety-first deny lists.
Claude Code Ultra Plan multi-agent workflow, Y Combinator AI-native startup strategy, RunPod serverless GPU deployment, OpenScreen, self.md plugin, and HarnessKit comparison — a survey of the AI development ecosystem
A comparison of open-source AI image processing tools from matting to upscaling — MODNet, ViTMatte, Diffusion-Sharpening, ImageSharpening-KD, and Upscayl — plus LINE emoji format constraints
Fixed four simultaneous bugs in the reranking pipeline, overhauled image generation from single-tone to dual-batch (3-tone/5-tone), and shipped a batch of frontend components including reactions, likes, and feedback.
Building a hybrid background removal pipeline that auto-runs rembg then lets you touch up with SAM2 click-based erase/restore, plus a wan2.6-i2v-flash 720P video upgrade and physics-based motion prompt rewrite.
The agent activity monitor was buried in a secondary tab while the chat interface dominated the screen. This post covers the full redesign using Tailwind CSS v4 and shadcn/ui to fix that structural problem.
Combining Claude Code with Figma to maintain consistent UI through design tokens and automate content design workflows, with a look at web accessibility automation possibilities.
Adding a multi-agent team design skill to HarnessKit, documenting agent design patterns, implementing CLAUDE.md auto-registration, and releasing v0.4.0
Adjusting Gemini API timeouts, fixing OTel local errors, improving tone/angle injection regeneration UX, scaling up EC2 instances, and adding deployment scripts
A data-driven analysis of Cole Medins video on AI replacing software engineers with added perspective from daily AI coding tool usage
