Low-cognitive first draft

A household opens a meal planner and sees an empty calendar. Seven blank slots, a search box, a prompt to "add recipes." Before dinner exists, someone must invent structure. How many meals, which constraints, what variety means this week. The product waited. The human did all the thinking.

Intelligent software can invert that opening move. Given the originating sentence, "plan dinners for four, Tuesday and Thursday busy, use the spinach before Wednesday," the Meal Companion can return a full provisional week. Meals assigned, busy evenings flagged, spinach placement visible, assumptions stated. The household's job becomes correction, not invention from zero.

Helpful context: Consumers direct intelligence sets the expectation that users guide outcomes. Point and fix and conversation becomes state are how correction stays local and durable.

Begin the work, do not finish it alone

A useful first draft is complete enough to react to and provisional enough to change. It exposes assumptions that materially shape the result: busy evenings treated as under twenty-five minutes of active effort, allergies enforced in code, spinach deadline honoured. The household can see what the product inferred and disagree with any of it.

A confident final that resists correction fails the principle. The draft is a collaborator, not a trap. Persistence before regeneration matters here because when Tuesday's meal is wrong, repair stays local instead of regenerating everything already accepted. Overconfident openings read like finals, and households experience correction as defeat. Useful openings read like invitations, and households experience correction as collaboration.

The difference in cognitive effort is substantial. In a blank-page flow, the household interprets the request mentally, searches and compares recipes, assigns across seven days manually, discovers constraint conflicts late, and rebuilds on first rejection. In a draft-first flow, the request compiles to working state, candidates generate and compose, a provisional plan returns, critique surfaces weaknesses early, and point and fix handles one slot at a time. Cognitive effort drops when the product supplies structure the household can react to. Fluent prose with no plan object behind it does not count.

Blank-page flow	Low-cognitive first draft
Interpret request mentally	Request interpreted into working state
Search and compare recipes	Candidates generated and composed
Assign across seven days manually	Provisional plan returned
Discover constraint conflicts late	Critique surfaces weaknesses early
Rebuild on first rejection	Point and fix on one slot

The first responsibility of intelligence is therefore not to impress users with sophisticated capabilities. It is to reduce unnecessary thinking.

Thoughtware: Designing in the Intelligence Age · Ch. 5

Restraint in the opening move

Designing effective first drafts requires restraint. The system produces enough structure to think with and enough humility to change. When authority or knowledge is missing, an unfamiliar medical diet, unclear leftover policy, the honest draft escalates with a targeted question rather than inventing a finished answer.

Critique before presentation helps. CritiquePlan exists so weaknesses surface in the draft, not at dinner time. A first draft that hides busy-day conflicts until Wednesday trains distrust. A first draft that names conflicts and offers patches trains collaboration. The opening move includes critique visibility: households see that Tuesday was flagged before they tap Tuesday. That is low-cognitive because it removes discovery effort, not because it hides problems.

Assumption visibility connects to visible thinking. Household size, busy evening treatment, allergy enforcement, spinach deadline, and repeat avoidance appear as readable assumptions when they bind the plan. Hidden assumptions turn correction into archaeology, and the first draft is the first place assumptions appear. If they are invisible there, every later surface fights an uphill trust battle.

When the product cannot finish honestly, it stops. Unfamiliar medical diet, missing authority for purchase, or unsettled guest count produce targeted questions, not a confident seven-day plan. An honest partial draft beats a fluent wrong draft. Households respect products that stop at the frontier of knowledge rather than inventing certainty they cannot defend.

What the product needs underneath

The opening move needs architecture. InterpretWeek compiles intent. GenerateCandidates produces meals. ComposeWeek assembles the plan. CritiquePlan surfaces weaknesses before the household discovers them at dinner time. Working state holds the provisional plan as an object. Evaluation can grade draft quality offline.

Products that return drafts without critique paths ship surprise. Products that critique without repair paths ship anxiety. The first draft sits in a loop with recommendation and local patch. After CritiquePlan runs, recommend, don't only respond proposes substitutions before the household asks. That sequence turns the first draft into a living artifact: the household receives structure, sees weaknesses named, and gets repair options with trade-offs. Idle chat asking "anything else?" wastes the opening move.

The pattern transfers wherever blank pages used to wait for human invention. An invoice queue returns a proposed routing and exception summary, not an empty form. A support tool returns a draft reply with cited sources, not a blank textarea. A contract tool returns a redline draft from a dispute summary. The clerk, support agent, or lawyer still reviews, patches, and approves. The product removed invention from zero, not responsibility.

Measuring the opening move

Time-to-first-correctable structure beats time-to-first-message. Households reach a plan they can patch within one interaction when terrain allows. Tracking how often users accept the draft without edit versus patch once versus regenerate all reveals the health of the opening move. Rising regenerate-all rates signal overconfidence or missing locality.

Draft quality is judged by correction cost, not fluency. Grading first drafts offline on constraint satisfaction, assumption visibility, and repair readiness catches problems that fluency benchmarks miss. A busy-week held-out set catches drafts that sound fine and fail Tuesday. Product analytics that segment users by draft acceptance path reveal collaboration health: patchers signal healthy collaboration, serial regenerators signal overconfidence or missing locality.

Building draft-first flows costs more than blank-page forms. The product needs InterpretWeek, GenerateCandidates, ComposeWeek, CritiquePlan, working state, and eval suites. The return is lower cognitive load for users and higher retention when correction stays cheap.

Common mistakes

Theatrical fluency without a plan object behind the text is the most visible failure. The product writes beautifully about a plan that does not exist as a structured object, and the household cannot patch what has no schema.

Overconfident finals that resist correction are equally damaging. The plan arrives looking finished, and the household treats correction as confrontation rather than collaboration. Invitation copy like "provisional week for four, busy Tue and Thu noted, spinach deadline Wednesday, tap any day to adjust" sets the right expectation. Final-sounding copy sets the wrong one.

Infinite chat openings ask the household to co-author structure turn by turn. The approach feels conversational but costs more cognitive effort than draft-first when terrain is familiar. And blank chat that never returns structure forces the household back to the same invention-from-zero experience they expected the product to remove.

Accepted plans produce downstream artifacts without manual re-entry. The first draft opens meal structure, deterministic code assembles the shopping list, and end-to-end low cognitive load spans multiple artifacts. When the downstream chain breaks, the opening move loses credibility regardless of how good the initial plan looked.

What to do next

The first artifact a product returns after intent capture determines whether the relationship starts as collaboration or as blank-page fatigue. Measuring time-to-first-correctable structure against a draft-first alternative reveals whether the opening move reduces cognitive burden or merely relocates it. The connection to correction UX is direct: point and fix makes the draft worth producing by keeping local repair cheap, and recommend, don't only respond turns the draft into an active collaborator rather than a static proposal.

See visible thinking and persistence before regeneration.

Read next: Point and fix.