Your AI First Draft Is Pulling You Toward Generic. Here's How to Avoid It.

Your AI first draft is not the output. It is a gravitational center — one that your team will anchor to, edit toward, and defend as a direction. Dylan Field, Figma's CEO, named this plainly in a Stratechery interview on June 25: AI output "draws from the middle of the distribution." The moment you paste it into a shared doc, it becomes the starting assumption.

What Field actually said

Field was talking about product design, but the principle applies to every role using AI to draft. He described employees becoming "viscerally attached" to a single AI-generated concept, which blocks the broader exploration that produces differentiated work. His phrase for where real value lives: "out of distribution by definition" — meaning the work that is genuinely distinctive is, almost by definition, not what the AI produces by default.

This is not a critique of AI drafts. It is a description of what happens when you stop at one. The middle of the distribution is competent. It reads like something that could work. That is precisely why teams stop iterating.

What to do differently on Monday

Treat the first draft as something to disagree with, not refine. After reading it, write down two things that are wrong for your specific situation — not generic quality issues, but things that are off for your audience, your context, your constraints. If you cannot name two, the draft is doing your thinking for you.

Generate at least three versions before evaluating. Field's concern was that teams anchor to a single concept too early. Ask the AI to produce a deliberately different version: different structure, different angle, completely different emphasis. Then compare, rather than incrementally editing the first one.

Separate generating from evaluating. Do not assess quality while the AI is producing. Run the generation, close the tab, come back with fresh eyes. The pull toward "this is pretty good, let's work with this" is strongest in the first two minutes after reading.

When good enough becomes the ceiling

Sarah is a content strategist at a 30-person SaaS company. She uses AI to draft email newsletters briefed from a template she built over two years — it works well and her open rates are strong. Last quarter, she looked back at three consecutive newsletters and noticed they all had the same structure, the same three-act pacing, the same type of CTA. The AI had a format. She had been editing within it.

She now generates two drafts from the same brief: one following the template, one with an explicit instruction to take a completely different structural approach. The second draft rarely ships. But it consistently surfaces at least one element the first one missed — a lead angle, a more specific example, a reframe — that makes the final newsletter more distinctive than what she would have edited from the first draft alone.

When the team stops ranging

Marcus is a brand strategist at a 200-person consumer goods company. His team runs three AI-assisted concept sessions per week — product naming, campaign angles, messaging territory. The process: brief in, AI out, team reacts and builds. After six months he noticed something. The brainstorm sessions had become faster. They were also producing fewer surprising concepts.

The AI had made them better at refining. It had made them worse at ranging.

He restructured the sessions. The first ten minutes use no AI: human-generated concepts only, intentionally rough, no self-editing. Then the AI session. Then comparison across both sets. The human-first concepts are usually looser. Some are weaker. The ones that win pitches are consistently from the human-first phase — or from what that phase surfaced that the AI session then developed.

The one-sentence version

Competent-and-coherent is exactly where teams stop exploring — and the distinctive work lives one or two moves past that.