Guides
How to build the parts of an AI product people see: streaming answers, tool calls, approvals, reasoning and agent runs.
- How to add a human approval step before an AI agent actsPause the agent before any action that sends, pays, deletes or deploys, and show one card with what it will do, the values that matter, the risk, and Approve or Reject. · 3 min read
- How to show progress for a multi-step AI agent runShow an agent run as a vertical timeline of its steps, with a live timer on the active one, evidence on finished ones, and a waiting step that holds the approval. · 3 min read
- How to show an AI model's reasoning or thinking stateWhile a model reasons, show one quiet line with what it is doing and a timer, steps behind a fold, and fold it into "Thought for 12s" once the answer starts. · 3 min read
- How to show tool calls and their results in an AI chatShow each tool call as one quiet row (what it did in plain words, the tool, a timer) and keep the input and output JSON one click away, not in the message. · 3 min read
- How to stream LLM text in React without the layout jumpingTokens arrive in uneven bursts, so text that renders each chunk as it lands jumps and reflows. Reveal whole words at a steady pace and hold the half-word. · 4 min read
New guides by RSS.