Talk to Your AI Lecture While It's Being Made

Until now, starting an AI creation in Scholarly meant handing it over. You chose a source, pressed Create, and a progress page told you which stage it was in. If the plan was not quite what you wanted, you found out when the lecture was done, and the fix was to make another one.

From today, the progress page talks. The agent making your creation posts short notes as it goes, and you can answer them. Ask for a worked example, a shorter section, a different order, or just a question about what it is doing, and it reads your message between steps and replies on the same page.

The short version

• The agent keeps you posted. Once it has settled its plan it says what it is going to make. When it starts writing, choosing pictures, rendering or checking, it says so in one line.
• Ask for a change while it works. Type into the thread on the progress page. The agent reads it between steps, applies it when it can, and tells you what it actually changed.
• Chat can pass a request along. Tell chat "ask my running video to add a review quiz at the end" and it relays the request into that creation's thread for you.
• Home shows the latest note. A running creation's row on Home carries the agent's most recent update, so you can see where it is without opening the page.
• Up to 15 requests per creation. Enough to steer a lecture, and a limit that keeps every creation on one credit.

What it looks like

Start a video lecture from a chapter on photosynthesis. A minute or two in, the thread reads:

Here's the plan: five sections, from light-dependent reactions to the Calvin cycle, with a diagram of the chloroplast up front. Say so now if you want something different.

That is your moment. Reply with "Add a worked example on how much glucose a leaf makes in an hour" and the thread shows Sent, then Seen once the agent has read it. A little later:

Added the worked example to section 3 and two review questions after the summary. Writing the narration now.

The finished lecture has the example in it. No second credit, no second run.

What you can ask for

The thread is for the kind of change you would make to a draft:

• Content. Add an example, a definition, a comparison, or a short recap. Leave out a topic the source covers but you do not need.
• Shape. Keep a section shorter, put the summary first, spend more time on the hard part.
• Voice. Simpler words, a more formal register, the terms your course uses.
• Questions. "Which pages are you using for section 2?" gets an answer, not a change.

Some things a thread cannot do. It cannot make a lecture longer than your plan allows, switch the model, or change the sources after the agent has read them. When a request is outside what it can do, it says so briefly and keeps working.

Where it works

The thread appears on the progress page of agent-built AI video lectures, AI Slides, AI documents and Research. Podcast threads become available when the agent makes the episode's extras, after the episode script is written; that agent can adjust the extras, but cannot rewrite the episode. Flashcards, quizzes, infographics and mind maps do not have a thread yet.

Replies arrive between the agent's steps, so a change can take a moment to show while a long step like rendering finishes. If a creation stops and restarts, the thread pauses and picks up where it left off.

Try it

1. Open any PDF, recording or page and choose Video lecture, Slides or a document.
2. On the progress page, wait for the agent's first note, or type straight into Ask for a change.
3. Watch for Seen, then the reply explaining what the agent could change in the creation.

Or ask chat: "Tell my running video lecture to add two review questions at the end." Chat passes it on and the request shows in the thread marked From chat.

Open Home and start a creation to try it.