OpenAI API: design a custom voice from a text prompt, plus nine new GPT-Live voices
Prompt-to-voice is now available to all API customers: describe a voice in words, then reuse it in GPT-Live sessions. Creating the voice is free.
Stay one prompt ahead.
Short, practical, source-cited tips for ChatGPT, Claude, Gemini, Grok, Cursor, Midjourney and more. Updated regularly.
Latest AI news all news →
Meta releases Muse Spark Open weights: 816B MoE, 1M context, runs on two Mac Studios
Meta open-sourced the weights of Muse Spark Open, a multimodal coding and agent model it calls the most capable US open-weight model to date.
Gemini 4 Argon is now generally available, at intro pricing through Nov 30
Google's new frontier model for coding, knowledge work and cyber defense is open to API developers, enterprises, and Pro/Ultra subscribers in the Gemini app.
Featured
Claude Code: give it a pass/fail check and let it iterate
The Claude Code team's #1 tip: hand the agent a test, build or linter it can run, so it fixes its own mistakes before you see them.
Cursor project rules must be .mdc files, or they're ignored
Rules in .cursor/rules need the .mdc extension plus frontmatter. A plain .md file there is silently skipped; use AGENTS.md for plain markdown.
ChatGPT Projects: give each workstream its own instructions and memory
Project instructions override your global custom instructions inside that project, and project-only memory keeps context from leaking across.
Latest tips
Claude Code: give it a pass/fail check and let it iterate
The Claude Code team's #1 tip: hand the agent a test, build or linter it can run, so it fixes its own mistakes before you see them.
Grok API: send the encrypted reasoning back every turn
On xAI's Responses API, append every output item (including encrypted reasoning) to the next request. Dropping it hurts quality and cache hits.
Use plan mode before multi-file changes in Claude Code
Press Shift+Tab to enter plan mode: Claude reads and proposes a plan without editing files. Refine the plan, then let it execute.
Gemini: put your question after the long context, not before
With big documents or code, Google advises supplying all context first, then your specific instructions at the very end, bridged by a clear transition.
Cursor project rules must be .mdc files, or they're ignored
Rules in .cursor/rules need the .mdc extension plus frontmatter. A plain .md file there is silently skipped; use AGENTS.md for plain markdown.
ChatGPT Projects: give each workstream its own instructions and memory
Project instructions override your global custom instructions inside that project, and project-only memory keeps context from leaking across.
Keep CLAUDE.md short (under ~200 lines)
CLAUDE.md loads into every session. Bloated files get ignored, so keep only always-relevant facts and move procedures into skills.
Midjourney: keep a consistent look with --sref and --sw
Add --sref with an image URL or style code to reuse a look across images, and tune its strength with --sw (0-1000, default 100).
Use hooks for anything that must happen every time
Instructions are suggestions; hooks are guarantees. Use Claude Code hooks to auto-lint after edits or block dangerous commands.
OpenAI API: tune reasoning effort and verbosity separately
reasoning_effort controls how hard the model thinks; verbosity controls how long the answer is. Adjust them instead of rewriting prompts.
Customize Audio Overviews: Brief, Critique or Debate instead of the default
Google's notebook tool (formerly NotebookLM) can turn your sources into a 2-minute brief, a critique of your draft, or a debate, plus a custom focus prompt.
Gemini: include a few varied examples in your prompt
Google recommends always including few-shot examples. Clear examples can even replace instructions for format and tone.
Run 3-5 coding agents in parallel with git worktrees
The Claude Code team's biggest productivity unlock: several sessions at once, each in its own git worktree so edits never collide.
Delegate noisy side tasks to subagents
Codebase searches and log digs flood your context. A subagent does the work in isolation and hands back only the summary.
Structure complex Claude prompts with XML tags
Wrap instructions, documents and examples in descriptive XML tags so Claude never confuses your data with your instructions.
Keep Cursor rules minimal; move occasional workflows to skills
Every rule costs context in every chat. Cursor recommends short, specific rules (<500 lines) and skills for things you need only sometimes.
Grok API: set reasoning effort explicitly and leave sampling near 0.7/0.95
xAI recommends temperature 0.7 and top_p 0.95, explicit reasoning effort on every request, streaming, and long client timeouts.
Midjourney: dial literal vs. artistic with --stylize
--stylize (or --s) ranges 0-1000, default 100. Go low to follow your prompt closely, high for more artistic interpretation.
Treat production prompts like application code
OpenAI recommends versioning prompts in your repo, using typed variables, reviewing them in PRs and running evals on every change.
The three C's: concise, clear, consistent, plus a new chat per topic
Google's prompting framework: persona, task, context, format. Write concisely and clearly, reuse the same terms, and start fresh chats for new topics.