Manus output UX: agent runs, files & deliverables
Updated June 30, 2026
Manus optimizes for delegated work, not quick Q&A. The thread shows agent status (“Manus is working”), file edit cards, and structured deliverables when a task completes. Meeting minutes mode adds a dedicated capture surface before summarization; slides graduate to a full preview modal with filmstrip navigation.
Dedicated capture surface
What works
- Mode opens a focused recording UI instead of jumping straight to chat output.
- Copy promises automatic summary after recording, which sets expectation for post-processing.
- Consent line (“confirm you have consent from all parties”) addresses meeting compliance upfront.
- Timer shows 0:00 / 2:00:00 so users know duration limits before they start.
What we would push on
- Mobile app promo below the card competes with Start on desktop.
- Composer is disabled during setup. Users cannot add context until recording ends.
Business strategy
Meeting notes are a high-frequency job for knowledge workers. A purpose-built capture card signals Manus handles the full pipeline (record → transcribe → summarize), not just text generation.
Tradeoff
| Decision | Benefit | Cost |
|---|---|---|
| Mode-specific recording card with consent and time limit | Clear job scope; compliance copy built in | Extra step vs paste-a-transcript; mobile upsell clutter |
Takeaway
For capture-heavy modes, use a dedicated start surface with limits and consent, not a generic prompt.
Pattern: First Success Flow
Pattern: Input Mode Toggle
Live recording feedback
What works
- Waveform differentiates active capture from idle state, so users trust audio is flowing.
- Recognizing… status text explains latency between sound and summary.
- Finish is prominent; composer shows “Recording in progress. Edit after recording ends.”
What we would push on
- No pause, only Finish. Long meetings may need pause without ending session.
Business strategy
Visible capture feedback reduces “is it working?” anxiety during agent tasks that take minutes. The same pattern applies beyond audio, so users need proof the agent is acting.
Tradeoff
| Decision | Benefit | Cost |
|---|---|---|
| Live waveform + locked composer during recording | Trust during capture; clear end action | No mid-recording edits or pause |
Takeaway
During long-running capture, show live feedback and lock unrelated inputs until the phase completes.
Pattern: Voice Input
Pattern: Progressive Disclosure
Work-in-progress trace
What works
- Status line names the phase (“Generating meeting summary”) with time expectation.
- File edit card shows the artifact being touched, so users see filesystem-level work.
- Floating “Manus is working 0:08” chip above composer keeps status visible while scrolling.
- Message Manus stays available for mid-run steering without canceling the task.
What we would push on
- File card preview is collapsed by default. Users must expand to see diffs.
- No explicit cancel on the working chip.
Business strategy
Autonomous agents fail when users feel blind. Manus exposes file edits and elapsed time to justify credit spend and build trust that work is happening off-screen.
Tradeoff
| Decision | Benefit | Cost |
|---|---|---|
| File edit cards + floating working chip during agent runs | Transparent progress; steer via composer mid-run | Technical detail may intimidate; weak cancel affordance |
Takeaway
Show what the agent is touching (files, steps) and keep a persistent working indicator with elapsed time.
Pattern: Response Refinement
Pattern: Progressive Disclosure
Structured completion
What works
- Output uses document structure (Meeting Summary, Key Points), not a wall of prose.
- Completion chip with checkmark and 1/1 ties back to the original task intent.
- Thumbnail of source file beside the chip links input to output.
- Agent explains empty or low-signal input honestly (no fabricated meeting content).
What we would push on
- No inline export or copy actions visible on the summary block in this view.
Business strategy
Structured deliverables feel worth credits. Pairing the completion chip with the source artifact closes the loop so users trust Manus read their file, not hallucinated a meeting.
Tradeoff
| Decision | Benefit | Cost |
|---|---|---|
| Structured output + task completion chip with source thumbnail | Clear done state; input-output linkage | Export actions not prominent in-thread |
Takeaway
Mark tasks complete with a chip that references the original intent and show structured sections, not chat bubbles only.
Pattern: Response Refinement
Pattern: First Success Flow
Artifact preview modal
What works
- Slides open in a dedicated preview (Pixel), deliverable is visual, not markdown in chat.
- Filmstrip thumbnails map deck structure; current slide highlighted with blue border.
- Modal keeps chat context behind, so users evaluate the artifact as a product, not a message.
What we would push on
- Preview is view-first, edit or regenerate actions are not obvious in this capture.
- Modal title “Pixel” is opaque vs “Slide preview” for new users.
Business strategy
Deck output competes with Gamma and PowerPoint copilots. A filmstrip modal positions Manus slides as shippable artifacts, justifying mode-specific composer setup and higher credit tiers.
Tradeoff
| Decision | Benefit | Cost |
|---|---|---|
| Full-screen artifact modal with slide filmstrip | Visual proof of value; deck navigation in one view | Branded modal name; edit path unclear in preview |
Takeaway
Promote heavy deliverables to a preview workspace with navigation, keep chat for status, not the artifact itself.
Pattern: Response Refinement
Pattern: Progressive Disclosure
Steal this
- Mode-specific capture UI before agent work (meeting minutes)
- Live waveform and locked composer during recording
- “Manus is working” chip with elapsed time + file edit cards
- Task completion chip linking back to original intent (1/1)
- Structured section headings in agent output
- Slide deliverable in a filmstrip preview modal
Skip this
- Autonomous runs with no visible file or step trace
- Delivering slide decks as markdown in the chat stream only
- Recording modes without consent or duration limits
- Completion states with no link between input file and output
How others output, artifacts & refinement
Same job, different product bets, and what each tradeoff reveals.
ClaudeClaude keeps quick answers in-thread and promotes artifacts to a split pane; Manus centers file edits and task completion chips for autonomous runs.
Read teardownClaudeClaude artifacts open beside chat for code and apps; Manus uses modal preview (Pixel) for slide filmstrip review.
Read teardown
ChatGPTChatGPT spreads refinement across regenerate and Activity; Manus exposes agent working state and filesystem edits during the run.
Read teardownOriginal gallery pages: Agent execution & deliverables