What it does
Before executing, the agent writes a structured plan of what it's going to build. User can review / edit the plan before execution starts.
Evidence
- Replit - Core Upgrade Modal — listed as the first Core feature with the description: "Describe what you want to build and Agent will create a plan for you. Build task plans in the background to get more work done at once."
- Replit - Workspace Home — a
Plantoggle is visible on the prompt input (presumably triggers Plan Mode; enabling likely prompts for upgrade if on free tier).
Why it matters
💭
Plan Mode is a risk-reduction feature for autonomous agents. Without it, the user prompts → agent does stuff → finds out what was built. With it, user prompts → agent plans → user reviews → agent executes. Catches bad approaches early. Particularly valuable for larger changes where rework is expensive.
This pairs with Replit - Autonomous Build Mode (long-running autonomous execution) — you'd want to review the plan before letting the agent run for 30 minutes unsupervised.
Gating
Core-only. Free tier has only reactive (no-plan) agent.
Differentiators
- Anything.com has no explicit "plan first" mode visible. The base agent's reasoning blocks (see Anything.com - Agent Chat) are structured but in-line; not a separate explicit step the user approves before execution.
- Cursor has agents that plan-then-execute — so this is a recognized pattern; Replit is productizing it as a tier, Anything may be inlining it.
Sources
- !raw/replit/Task_View.png
- !raw/replit/Project_View.png