Draft-First LLM Augmentation
Never ask the model to do your job — write your version first, then have it improve it.
- Difficulty
- Starter
- Time to result
- ~days to results
- Steps
- 4
- Confidence
- 90%
Nika's discipline for using ChatGPT inside a PM workflow: you must arrive with your own thinking already formed, then use the model as an amplifier on three specific jobs — rewriting a mission statement for a multi-audience readership, generating user segments your own mind wouldn't reach, and proposing AI-enhanced feature ideas. The rule that separates augmentation from outsourcing is that you supply the intent and the first draft.
Origin
Marily Nika's daily working practice as a product lead at Meta, described in early 2023.
Core principles
- 01The model enhances your worth; it does not steal from you.
- 02You bring the mission and the intent; the model brings the phrasing and the breadth.
- 03A mission statement is read by leadership, junior staff, other departments and competitors — it must land for all of them, and that is a language problem an LLM is unusually good at.
- 04Reserve the model for the low-strategy, high-repetition parts of the job so you can spend your time on strategy.
How to run it
- 1
Form the intent yourself first
Decide what you actually want — the mission, the direction, the product intent — before opening the model. Write your own version.
Watch out Skipping this step is where PMs slide from being augmented to being replaced by their own workflow.
- 2
Ask for a rewrite, not a creation
Paste your draft mission statement and ask the model to rewrite it. Judge the output against the bar of 'would a kid understand this and still be inspired?', because it will be read by every discipline and seniority level.
Pro tip Nika finds even the first attempt typically beats her own draft, because the model optimises for the multi-audience reading, not the PM's insider context.
- 3
Mine it for user segments
Prompt in the form 'who would be interested in [product with specific constraint]?' — e.g. a fitness band that doesn't have a screen — and it returns a bulleted list of segments with their motivations and pain points, including ones you would never have reached.
Pro tip Put the awkward constraint into the prompt; the constraint is what forces non-obvious segments to surface.
- 4
Push the repetitive artefacts down to the model
Identify the sections that repeat across all your PRDs and projects and let the model draft those, freeing your time for the strategic side.
In the wild
Nika demonstrates segment generation by asking who would be interested in a fitness band that doesn't have a screen. The model returns segments like young professionals who are interested but short on time, and people who don't want to charge a wearable every day — each with motivations and pain points attached.
→ A richer segment list than the PM would have produced alone, generated in a single prompt.
Common mistakes
Asking the model to do the job instead of improve it
Nika is explicit that she isn't making it do her job for her — she asks only after she already has the mission in her head. Without an intent to anchor it, the output is generic and you have outsourced the judgment that is your actual value.
Fearing replacement rather than using the tool
Nika's view is that the technology enhances rather than steals; PMs who treat it as a threat forgo the time it would free up for strategic work.
Is it for you?
Best for
Product managers and operators writing mission statements, PRDs, and segment definitions who want leverage without outsourcing judgment.
Not ideal for
Work where the specific, non-public context is the entire value and a generic rewrite would strip it out.
From the transcript
“so what I do is I literally goild the CH and I say rewrite this mission statement for me”
“but I'm not making it do my job for me I'm asking it after I have already had a mission in my head and what…”
“so you would say something like who would be interested in a fitness band that doesn't have a screen and it will provide a bulleted…”
From the episode
AI and product management
Marily Nika (Meta, Google)