Read long agent replies in plain words
Plain Speak adds a short card to each agent reply: what you asked, the outcome, what to do next and any questions.
Agent replies can be long and technical. Plain Speak adds a short card to each reply, in the same order every time, so you find what matters without reading it all. It is off until you turn it on.
Turn it on#
Open Settings > Plain Speak.
Turn on the Plain Speak switch at the top of the page.
From then on, the agent writes a card with each new reply, in every session. Older messages stay as they are. The agent writes the card itself as part of its answer, so it costs nothing extra.
What the card shows#
Latest: a one-line recap of what you last asked for.
TLDR: the outcome or answer, in one line.
Recommended Action: the single most useful next step, like Answer questions, Archive or Wait.
Response: a few plain points backing up the outcome, when they help.
Questions: the questions the agent asked you, which you can answer from the card.
A part with nothing to say is left out.
Switch between the card and the original#
A reply opens on the agent's original text. A small button in the message's header shows which view you are on (Original or Plain Speak); click it to switch. Or press VVVV to flip the latest message.
To open replies on the card instead, turn on Show Plain Speak first. It changes only your own view.
Right-click the header button and choose Plain Speak settings to jump to these settings.
When there is no card#
A message shows only its original text when Plain Speak is off, when the agent wrote no card (for example in a session from before you turned it on), or when the message contains a phrase you listed under Skip Plain Speak for phrases.
Emojis on action labels. The same settings page can show a small emoji before each Recommended Action, so the next step stands out. It is on by default.
Quick replies can set Latest. A quick reply can put its own text in the Latest line of the reply it gets back.
Plain Speak and questions. The Questions part works like the question buttons under a message.
Related
- Choose how much agents do on their ownAgent Permission Level decides when a Claude agent stops to ask you, from asking before every change to never asking.
- Batch alerts with Focus ModeFocus Mode holds back "needs you" alerts and releases them together, after a few pile up or some time passes, so you can work without interruptions.