Better references beat better prompts
Most people skip the crafty part, the initial context. Research papers, niche blog posts, specific docs, reference implementations. Feeding the right material in is the highest-value move you can make with AI coding tools.
Based on workflow patterns from building with Claude Code and Codex throughout 2025 and 2026.
The prompt matters less than what surrounds it
I built an ASCII art renderer (now part of fx.blode.co) in a day. Not off a clever prompt, but by starting from a blog article on ASCII rendering and handing the whole thing to the agent. That article was the context. Then I built allmd so I could turn any URL, PDF, video, or audio into markdown and feed it in the same way.
Most people refine their prompt for an hour when fifteen minutes finding the right reference would do more. Better references beat better prompts.
Front-load the right material
My Raycast snippet "zcc" expands to: "Do extensive research. Make a plan with phases and todos. Use a swarm of subagents and teams." Even that is just research before you build. The best result always comes from the attempt where I front-loaded the right material, never the first one.
I run Brian Lovin's /simplify skill at the end of every session, so the agent grades its own output.
The people who get the most out of these tools curate inputs instead of writing sophisticated prompts. The difference is almost never the model. It's the fifteen minutes you did or didn't spend gathering context before you hit enter.