Iterating with a production-grade agent burns real tokens and produces real side effects, even when the user is just exploring.
← Back to patterns
Memory, Context & Boundary
Draft Mode
Two modes, two trust zones
Split AI work into two explicit modes. In Draft: a cheap or no-side-effects pass, ideal for trying ten ideas in the time of one. In Live: the production model, real outputs, real cost. The 2026 industry has settled on "Plan Mode" terminology (Cursor, Claude Code) — the underlying move is the same: a deliberate, visible promotion from sketch to commit.
Metaphor
Sketch Pad
Interaction
Mode Toggle, Promote-to-Live
Use case
Exploration, Iteration, Pre-commit Workflows
Failure mode
Uniform Cost
Control mechanisms
Draft / Live Switch · Cheap-Model Badge · Sandbox Banner · Promote Button
Reference examples
Midjourney Draft Mode · Cursor Plan Mode · Claude Code Plan Mode · Replit Agent Sandbox
Concept Frame
Exploration deserves its own tier.
Most AI products quietly punish iteration. Every "let me try one more" runs the production model with full token cost and full side effects, even though the user is still figuring out what they want. Draft Mode names that asymmetry — the cheap tier is loud about being cheap, the live tier is loud about being committed, and promotion is one deliberate move rather than a hidden upgrade.
Interactive Example
The live side-by-side is the comparison.
This prototype already contains the contrast between chat and the pattern response. It stands as the primary comparison surface for this page.
Every turn costs the same in chat. Exploration shouldn't.
Iterating with a production-grade agent burns real tokens and triggers real side effects, even when the user is only trying ten versions of one idea. Chat flattens every turn to the same model, the same cost, the same persistence, so there's no cheap place to think out loud and leave the mess behind. Exploration is taxed at the price of commitment.
Draft Mode splits the work into two trust zones: a cheap, sandboxed Draft for trying ten ideas in the time of one, and a committed Live for the real output. The mode is always visible, the cost is deliberately asymmetric, and promotion from sketch to commit is an explicit act rather than a side effect of typing. The industry settled on "Plan Mode" terminology, but the move is older than the label.
"Give exploration a sandbox with a visible banner, and commitment becomes a choice instead of an accident."
DOPE Lens
How this pattern improves AI interaction quality.
Observability
Mode is always visible; sandbox banner makes the trust zone unmissable
Directability
Iterate as many times as needed in Draft; promote a chosen one to Live
Predictability
Draft outputs are explicitly marked as cheap-model approximations
Explainability
Each mode declares its model, cost profile, and whether side effects apply
Fit
Where it earns its place.
Use it
when
when
Iteration is exploratory and the production pipeline is too expensive to use as a sketchpad.
Avoid
when
when
The task is short and a single-tier model already serves both exploration and commit cheaply.
In the wild
Where this pattern already ships.
Real implementations from products users already know. Each entry shows how the pattern got applied — what shipped, where it sits in the product, what it traded off.
Midjourney · Draft Mode
Midjourney
The clearest example of the cost-asymmetric half, and the product the pattern is effectively named after. In V8.1 and V8.2 one draft prompt returns 24 images at 512px for 0.4 GPU minutes; in V7 a draft costs half a standard job. Promotion is a deliberate act on one chosen image — Vary or Remix re-renders it at SD or HD (Enhance in V7). The coding Plan Modes below ship the sandbox and the promotion; only this one makes exploring genuinely cheaper.
Visit Midjourney ↗
Cursor Plan Mode
Cursor
Shift+Tab toggles a no-edit planning mode where Cursor proposes a plan but cannot write files; user explicitly promotes to Agent mode for execution. Strongest 2026 example.
Visit Cursor ↗
Claude Code Plan Mode
Anthropic
Shift+Tab toggles a no-edit planning mode in the terminal. Claude proposes a plan; user promotes to execute. Same paradigm as Cursor — the 2026 industry settled on Plan vs Agent terminology.
Visit Anthropic ↗
Replit Agent Sandbox
Replit
Caveat — captures the side-effect-free preview half but not the cheap-model half. Sandbox-to-Deploy is the explicit promotion act; Effort-Based Pricing makes cost asymmetric.
Visit Replit ↗
Research Frame
Why this pattern exists, and where it breaks.
Offer two explicit modes: a cheap, sandboxed Draft for exploration and a committed Live for the final pass. Make the mode visible, the cost asymmetric, and promotion a deliberate act.
Chat treats every turn the same — same model, same cost, same persistence. There is no surface for cheap exploration that the user can leave behind.
If Draft outputs look identical to Live, users may trust them as final; weak mode signalling can blur the boundary.
Common questions
What people ask about this pattern.
When should I use the Draft Mode pattern?
+
Iteration is exploratory and the production pipeline is too expensive to use as a sketchpad.
When should I avoid the Draft Mode pattern?
+
The task is short and a single-tier model already serves both exploration and commit cheaply.
What problem does Draft Mode solve?
+
Iterating with a production-grade agent burns real tokens and produces real side effects, even when the user is just exploring.
Why is chat the wrong fit for this?
+
Chat treats every turn the same — same model, same cost, same persistence. There is no surface for cheap exploration that the user can leave behind.
Related Patterns
Curated connections, not just neighbors.
Hand-picked relationships — what this pattern pairs with in practice, what it replaces, what it builds on, and what builds on it.
Often paired with
Compute Budget
Draft Mode keeps exploration cheap and side-effect free; Compute Budget caps what the Live run can spend.
Often paired with
Parallel Alternatives
Run Parallel Alternatives in Draft, where they cost little, then promote the one you pick to Live.
Alternative to
Intent Preview
Draft Mode runs a cheap sandboxed pass to show a rough result; Intent Preview lists the steps and risks before anything runs.