@evan-dong

A Practical Seedance 2.0 Prompt Structure for More Controllable AI Video

3d ago · 4 min read · I have been collecting Seedance 2.0 prompt examples, and the pattern is clear: the best prompts are not longer. They are more structured. We published the collection as an open GitHub repo: https://gi

Testing Claude Code Dynamic Workflows With a 111-Agent Research Run

4d ago · 6 min read · Claude Code dynamic workflows are easy to misread as a "spawn more agents" feature. That is not the most useful way to think about them. The useful part is that a workflow turns a long task into a vis

Gemini 3.5 Flash vs Claude Opus 4.7 vs GPT-5.5: Which Model for Which Agent Task (May 2026)

May 20 · 5 min read · Google shipped Gemini 3.5 Flash at I/O 2026 on May 19. It outscores its own flagship Gemini 3.1 Pro on coding and agent benchmarks, runs 4x faster, and costs 40% less. But it does not win on every tas

Claude Code HTML Artifacts: When Markdown Is the Wrong Output

May 9 · 2 min read · Markdown is still the right format for many developer docs. But when Claude Code is generating a spec, review, report, or design comparison, Markdown is often the least useful final form. The Claude Code team has been experimenting with a simple shif...

Building with OpenAI's New Realtime Voice Stack: Architecture Decisions and Trade-offs

May 8 · 3 min read · OpenAI shipped three realtime voice models this week. If you're building voice-powered products, here's how to think about the architecture. The Stack ModelPurposeConnectionPricing GPT-Realtime-2Voice conversation + reasoningWebRTC/WebSocket/S...