Surface
An active agent conversation. Screenshot captures agent mid-work on prompt "Build me a calculator".
Layout
Single centered chat column (no split preview in this capture). Icon-only left rail visible at ~56px.
Conversation components
User message
- Component: rounded rectangle, light grey background, right-aligned.
- Content:
"Build me a calculator". - Typography: sans, ~15px.
Agent reasoning block
This is the key feature. Before generating code, the agent shows named reasoning stages:
Section 1 — "Considering Platform Implementation" - Prefix: Anything logo icon + small spark/sparkle icon (signals "thinking"). - Bold section name, followed by prose:
"available context, it seems appropriate to plan for both platforms, given the /apps/web endpoint present. Next, I need to consider component reuse."
Section 2 — "Prioritizing Web Implementation" - Same format: icon + heading + prose:
"I've decided to prioritize a web implementation initially, since there are no instructions on which platform to use. I'm focusing on the 'High-Fidelity SaaS' design system now, considering the available color palettes, typography, and specific components like feature cards and pills. SaaS buttons will be essential, especially 'Action Blue' for the primary actions."
What this reveals
The agent internally references a design-system taxonomy:
- Named design systems: "High-Fidelity SaaS", "Ghost Structure" (seen in later captures).
- Named component types: feature cards, pills, SaaS buttons.
- Named tokens: "Action Blue" for primary action color.
Anything.com's agent appears to have a curated catalog of design-system presets (selectable via the Default pill on the prompt — see Anything.com - Workspace and New Project). When the agent works, it narrates which preset it's applying. This is the system's equivalent of a "style guide" and is a product surface worth a dedicated page: Anything.com - Named Design Systems.
Prompt box during generation
Bottom of screen:
- Placeholder changes to Make updates to your app… (implying follow-up iterations, not new projects).
- Toolbar (identical to initial prompt): +, image, Figma, Auto >, Default >.
- Additional CTA inline: "Try MAX [New]" — upsell even while the base agent is running.
- Submit button becomes a square STOP button (circle with filled square) while agent is in-flight.
Pattern notes
- Agent Reasoning Blocks — named, human-readable stages. Not a raw chain-of-thought dump. Roughly 2-4 sentences per stage.
- Stop-during-generation — submit affordance swaps to stop. Clean pattern.
- No typing indicator / spinner visible — the reasoning blocks themselves signal activity.
Components observed
| Component | Purpose |
|---|---|
| Right-aligned message pill | User input |
| Icon + bold heading block | Agent reasoning stage |
| Prose paragraph under heading | Reasoning detail |
| Modified placeholder ("Make updates…") | Contextual prompt |
| Stop button | Halt generation |
| "Try MAX" inline promo | Upsell during base-agent work |
Sources
- !Chat_View.png