Capstone: Mock Loop & Self-Exam
Run the full loop on yourself before they do
Mock practical: a real launch exercise
After this you can After this you can produce a launch/positioning artifact under realistic constraints and defend every choice.
Cursor's loop almost certainly hands you a take-home or live exercise, because the team runs on paid trial projects and wants to see the work, not hear about it. Treat the mock as the real thing: pick a feature, ship the artifact and be ready to defend it line by line.
The trap is producing a polished deck that says nothing a developer would believe. The bar is a differentiated positioning backed by an insight, copy a working engineer wouldn't roll their eyes at and a distribution plan that names channels and metrics. Build all three in one sitting under a clock.
The prompt to run on yourselfPick one feature; ship three artifacts
Choose a real or hypothetical Cursor capability and produce a positioning doc, a launch blog and a channel plan with a metrics layer. Use a feature you actually understand, because the panel will ask how it works.
ICP, the competitive alternative, the one differentiated value and the proof.
April Dunford shape: who it's for, what it's against, why it wins and the evidence a skeptic accepts.
The actual copy, headline through CTA, written in a developer-credible voice.
Show the workflow, not adjectives. A real before/after a working engineer recognizes.
Sequenced across X, Hacker News, GitHub and YouTube with owners and timing.
Each channel gets a tailored asset and a metric, not one blast reposted everywhere.
Run it in five timed moves90 minutes if take-home; 45 if live
- 1Pick the feature and the ICP (10 min). Name a real segment. "Senior backend engineers on large monorepos" beats "developers." Write the job-to-be-done in one sentence.
- 2Find the insight (10 min). State the one true thing about the user or market that the positioning hangs on. Without an insight you have features, not a story.
- 3Write the positioning (15 min). Competitive alternative, the single differentiated value and the proof a skeptic accepts. One sentence of positioning, not a paragraph of hedges.
- 4Draft the launch copy (35 min). Headline, lede, the workflow shown concretely and a CTA. Cut every claim you can't show.
- 5Build the channel + metrics plan (15 min). Per-channel asset, sequence, owner and the activation or conversion metric you'd watch.
Interactive diagram. Tab through its regions; each focused region shows its detail in the panel below.
Positioning is the gate: nothing downstream is credible until it's sharp.
Submit the reasoning above the copy: a short "how I got here" that names the ICP, the insight and why you positioned against this specific alternative. Reviewers grade judgment as hard as output and a clean rationale lets a hiring manager trust the choices even where they'd have chosen differently. The artifact proves you can write; the rationale proves you can think.
Use AI with visible judgment
Cursor screens for authentic AI usage, so using AI is expected and pasting raw output is disqualifying. Show the seams: where you prompted, what you kept and what you rewrote because the model's version sounded like marketing.
- Move
- Drafting
- What good looks like
- AI for a first skeleton, then a heavy human rewrite in your own voice
- The tell that sinks you
- Shipped copy that still has the model's cadence and empty adjectives
- Move
- Research
- What good looks like
- AI to surface competitor claims, then you verify each against the real product
- The tell that sinks you
- Quoting a competitor capability the model hallucinated
- Move
- Defense
- What good looks like
- You can explain why each kept line survived and each cut line didn't
- The tell that sinks you
- "The AI wrote that part" when asked to justify a sentence
| Move | What good looks like | The tell that sinks you |
|---|---|---|
| Drafting | AI for a first skeleton, then a heavy human rewrite in your own voice | Shipped copy that still has the model's cadence and empty adjectives |
| Research | AI to surface competitor claims, then you verify each against the real product | Quoting a competitor capability the model hallucinated |
| Defense | You can explain why each kept line survived and each cut line didn't | "The AI wrote that part" when asked to justify a sentence |
Be ready to point at any sentence and say why it's there.
Self-check against the JD
- Positioning is differentiated, not a feature list dressed as benefits.
- The voice survives a developer's BS detector: specific, no hype, shows the workflow.
- The plan is end-to-end GTM, not just copy: pre-launch validation, launch and a post-launch metric.
- Every claim is one you could defend live to a skeptical DevRel peer.
The most common failure is positioning against "writing code by hand" instead of against the real competitive alternative, which is GitHub Copilot, Windsurf or Claude Code. Developers already use an AI tool; vague positioning that ignores the incumbent reads as a marketer who hasn't studied the market. Name the alternative honestly and win on a specific axis, like codebase-wide context or agent multi-file edits, rather than claiming Cursor is best at everything.
Takeaway. Ship all three artifacts - positioning, copy and a channel/metrics plan - and submit the reasoning above the work so judgment is as visible as output.
Self-check
QFor the mock launch exercise, which positioning frame is strongest for a Cursor feature aimed at senior engineers?
Mock demo-back
After this you can After this you can rehearse the Cursor demo-back round to time, narrating value over clicking.
Every GTM hire at Cursor demos the product back during the loop and the panel is full of engineers who use Cursor daily. This is the round where fake fluency dies. You have to drive the product like a power user and narrate why a developer should care.
The mistake is a feature tour: clicking around while the panel waits for a point. Lead with the job-to-be-done, reach the moment the value lands and let the narrative carry the demo while your hands keep up. The clicking is evidence; the story is the pitch.
Structure a 5-10 minute demoOne workflow, end to end
- 1Open with the job-to-be-done (30s). Name the developer and the task: "A backend engineer inheriting an unfamiliar service needs to ship a fix without reading the whole repo." No product yet.
- 2Set the scene fast (1 min). Open a real repo, not a toy. Show the friction in the old way for one beat so the contrast is felt.
- 3Drive to the aha (3-4 min). Run the actual workflow - Agent across files, codebase-aware ⌘K, Tab in flow - and narrate the value at each step, not the mechanics.
- 4Land the value out loud (1 min). Say what just changed for the engineer: minutes instead of an afternoon, a real PR in week one.
- 5Handle one objection live (1-2 min). Invite the skeptic's question and answer it honestly, including what Cursor doesn't do.
Interactive diagram. Tab through its regions; each focused region shows its detail in the panel below.
Five beats inside the window - the narrative leads, the clicking is evidence.
"I'm not going to narrate every keystroke. The thing to watch is that I never opened a file to find context - the agent pulled the right three modules itself and that's the difference between an autocomplete and something that actually understands your codebase. Here's what that saves a new hire on day one."
Make the narrative lead, not the clicking
- Speak the value before you act, so the panel knows what to watch for.
- If a step is slow, keep talking - dead air while a model thinks reads as a stall.
- Cut anything that doesn't move the job-to-be-done forward, even if it's a cool feature.
- End on the outcome for the developer, not on the screen.
Handle the skeptic honestly
Someone will push: "How is this different from Copilot?" or "Doesn't it hallucinate on a big repo?" Treat the objection as a gift. Concede what's true, then show the specific thing Cursor does better.
- Objection
- "This is just Copilot."
- Weak answer
- "No, Cursor is way better at everything."
- Credible answer
- "Copilot completes lines well. Watch what happens when the change spans four files - that's where the agent and codebase context separate."
- Objection
- "It'll hallucinate on our monorepo."
- Weak answer
- "It's very accurate, don't worry."
- Credible answer
- "It will sometimes. That's why the review step matters - here's how I'd verify a multi-file diff before accepting it."
- Objection
- "My team won't change tools."
- Weak answer
- "Everyone switches eventually."
- Credible answer
- "Adoption is the real risk. The wedge is Tab - it earns trust in a day without changing how anyone works."
| Objection | Weak answer | Credible answer |
|---|---|---|
| "This is just Copilot." | "No, Cursor is way better at everything." | "Copilot completes lines well. Watch what happens when the change spans four files - that's where the agent and codebase context separate." |
| "It'll hallucinate on our monorepo." | "It's very accurate, don't worry." | "It will sometimes. That's why the review step matters - here's how I'd verify a multi-file diff before accepting it." |
| "My team won't change tools." | "Everyone switches eventually." | "Adoption is the real risk. The wedge is Tab - it earns trust in a day without changing how anyone works." |
Honesty about limits reads as product fluency, not weakness.
Record the run and watch it back with a stopwatch. Mark every "um," every "so let me just," and every second where your hands led and your mouth lagged. Cut filler, tighten the open and confirm the whole thing lands inside the window with the objection still answered. Re-run until the narrative is the same crisp shape every time, because under panel pressure you default to your rehearsed version.
Have a backup workflow staged in a second window. Live demos stall: a model times out, a repo won't load, an extension misbehaves. If your only workflow breaks, you're improvising in front of engineers. Pick a second job-to-be-done you can run cold, keep the repo open and ready and rehearse the pivot so a stall costs you ten seconds, not the round.
Takeaway. Open with the job-to-be-done, let the narrative lead the clicking, answer the skeptic honestly and keep a backup workflow staged.
Self-check
Mock behavioral & founder round
After this you can After this you can pressure-test your stories and market POV out loud and update your view under challenge.
The founder round at a flat, talent-dense company isn't a culture chat. It's a probe for truth-seeking, taste and whether you actually have a view. They'll push on your stories and on your read of the market and the update under pressure is the signal.
Run this as a rapid-fire drill with a partner, on camera. The goal is concise STAR delivery with real numbers and a market POV you can defend and revise without going mushy. A view you abandon at the first push is as weak as one you cling to past the evidence.
The four stories to have coldEach in under two minutes, with numbers
One true personal moment, not the careers-page mission.
Land on: I want to market a dev tool engineers already love and I use it myself.
A launch you ran idea-to-iteration with little scaffolding.
Show what you wrote and shipped yourself, plus the metric it moved.
A time you disagreed with product or eng and were right to.
Show the data or insight, the respectful disagreement and the outcome.
A view you updated when the evidence turned.
This is the truth-seeking test; a blank here gets caught in the founder round.
When you tell the pushback story, name the specific thing you'd have lost if you'd stayed quiet and credit the person who changed your mind in the changed-my-mind story. Cursor screens for thoughtful disagreement and intellectual honesty at once, so a candidate who can both hold a line on evidence and concede gracefully reads as a real collaborator. Avoid the fake-flaw version where your changed-my-mind story is secretly a humblebrag.
Defend a market POV, then update itThe truth-seeking core
Have a real, specific take on the AI code-editor market and rehearse defending it. Then have a partner challenge it with a fair counter and practice updating without collapsing. The move is to absorb the new evidence and restate a sharper position.
- 1State the POV in one sentence. "The moat in AI editors is context quality, not model access, because every tool can call the same models."
- 2Back it with evidence. A specific workflow, a competitor's gap or a usage pattern you've seen - not a vibe.
- 3Take the challenge. Let your partner argue the counter: "Models are improving so fast that context advantages get commoditized."
- 4Update visibly. Concede the valid part, then sharpen: "Fair on raw capability. The durable edge is the data flywheel from how people actually use the editor, which a model release doesn't hand a competitor."
"You're right that the model gap narrows every quarter - I'd update my earlier claim. What I don't think commoditizes is the product surface and the usage data behind it. So my sharper view is that the editor, not the model, is where this is won."
Three questions that show depth
- A product question that reveals you've used Cursor hard: "How do you think about the trust gap on accepting large agent diffs?"
- A market question with a real POV underneath: "As enterprise becomes a bigger motion, how do you keep the developer-credible voice that won the bottom-up audience?"
- A role question about ownership: "Where does PMM end and DevRel begin here and what's the most contested seam?"
Rambling kills the behavioral round. A two-minute STAR story that wanders into four minutes signals you can't edit your own thinking, which is the opposite of the writing craft the role needs. On playback, cut the setup, lead with the situation in one line and get to your action and the number fast. The same discipline that makes a launch blog tight makes a story land.
Takeaway. Have four stories cold with numbers and treat a market challenge as a chance to update visibly rather than defend a frozen view.
Self-check
QIn the founder round you state a market POV and the founder pushes back with a fair counter-argument. What response best demonstrates the truth-seeking Cursor screens for?
Self-scoring rubric
After this you can After this you can grade yourself against a Cursor-flavored PMM bar with evidence from your mocks.
One honest scorecard beats another reread of the prep. Score yourself cold on the dimensions the loop actually tests, using evidence from your mock artifacts and recordings, not how you'd do on a good day.
Each dimension is 1-5. A 5 means a hiring manager could ship your artifact and trust your judgment without a rewrite. Anything below that means there's a named drill, which is the point of scoring before they do.
- Dimension
- Positioning rigor
- What a 5 looks like
- ICP, real competitive alternative, one differentiated value and proof a skeptic accepts
- Your score
- __
- Dimension
- Writing / narrative craft
- What a 5 looks like
- Copy that shows the workflow, survives a developer's BS detector and has no filler
- Your score
- __
- Dimension
- Product fluency
- What a 5 looks like
- You demo Cursor like a power user and explain Tab, Agent, ⌘K and context unprompted
- Your score
- __
- Dimension
- GTM / launch execution
- What a 5 looks like
- End-to-end plan: pre-launch validation, sequenced channels, owners and timing
- Your score
- __
- Dimension
- Metrics literacy
- What a 5 looks like
- You name the activation/conversion metric per asset and how PMM moves the funnel
- Your score
- __
- Dimension
- Values fit
- What a 5 looks like
- Truth-seeking, high ownership and taste are visible in your stories and choices
- Your score
- __
| Dimension | What a 5 looks like | Your score |
|---|---|---|
| Positioning rigor | ICP, real competitive alternative, one differentiated value and proof a skeptic accepts | __ |
| Writing / narrative craft | Copy that shows the workflow, survives a developer's BS detector and has no filler | __ |
| Product fluency | You demo Cursor like a power user and explain Tab, Agent, ⌘K and context unprompted | __ |
| GTM / launch execution | End-to-end plan: pre-launch validation, sequenced channels, owners and timing | __ |
| Metrics literacy | You name the activation/conversion metric per asset and how PMM moves the funnel | __ |
| Values fit | Truth-seeking, high ownership and taste are visible in your stories and choices | __ |
Senior-IC bar: a 5 = ship-it-without-a-rewrite. Score what you can show today.
Interactive diagram. Tab through its regions; each focused region shows its detail in the panel below.
Each dimension maps to a stage - these are ranked by how decisive they are for a Cursor panel.
Anchor each score to evidenceNo score without a pointer
Don't score from memory. Point each number at a specific artifact line or a recording timestamp, so the score is defensible the way your launch exercise has to be.
- Positioning rigor
- The positioning sentence and the competitive-alternative line in your doc
- Writing craft
- The launch blog's headline and lede; count the empty adjectives
- Product fluency
- The demo recording: did the narrative lead, did you handle the objection cold
- GTM execution
- The channel plan: per-channel asset, owner, sequence and a metric each
- Metrics literacy
- Whether you named a baseline and a target, not just "track engagement"
- Values fit
- The changed-my-mind story and where you used AI with visible judgment
Your readiness is set by your weakest two dimensions, because the loop has a stage for each. Averaging a 5 in product fluency against a 2 in metrics literacy doesn't make you a 3.5 candidate; it makes you the candidate who fumbles the funnel question. Circle the two lowest scores, ignore the strong ones and spend your remaining prep entirely there.
Self-scoring inflation is the silent failure. The fix is to grade against a real artifact a hiring manager would actually ship, not against your intent. If you can't point to the differentiated value sentence in your own positioning doc, that dimension is not a 5 no matter how clearly you can describe what you meant to do. Score the output, not the ambition.
Takeaway. Score each dimension 1-5 with a pointer to real evidence, treat a 5 as ship-it-without-a-rewrite and target your two lowest scores.
Self-check
Final readiness checklist & gaps
After this you can After this you can convert your self-exam results into a focused action plan and a go/no-go bar.
The last step is turning scores into drills. Confirm the assets exist, close the gaps the rubric surfaced and re-time the two timed rounds until they fit comfortably. Walk in able to state Cursor's positioning and your market POV in under 60 seconds.
This is a packing list, not new study. If an asset is missing or below a 4, you have a specific exercise, not a reason to panic. The prep is finite and the bar is knowable.
Confirm the assets existAll five, polished and rehearsed
- A specific, personal "why Cursor" you can say cold in under 90 seconds.
- Three STAR stories with real numbers, mapped to ownership, pushback and a changed mind.
- A demo-back of one real Cursor workflow, timed to the round and with a backup staged.
- A reusable positioning template you can fill for any feature in minutes.
- A writing sample - your launch blog from the mock - that shows the developer-credible voice.
Close the gaps with named drillsOne drill per low dimension
- Positioning rigor
- Re-fill the template on three different features until the differentiated value comes fast
- Writing craft
- Rewrite the launch blog's lede five ways; cut every empty adjective, then read it aloud
- Product fluency
- Demo a real workflow on camera daily; build the honest side-by-side vs Copilot and Windsurf
- GTM execution
- Rebuild the channel plan with owners, sequence and one metric per asset
- Metrics literacy
- Attach a baseline and a target to every asset; rehearse how PMM moves activation and conversion
- Values fit
- Re-record the changed-my-mind story; rehearse updating a market POV under live challenge
Re-time the two timed rounds
- 1Demo-back. Run it to a stopwatch until it lands inside 5-10 minutes with the objection still answered and no filler.
- 2Practical. Re-run the launch exercise under the clock until the three artifacts and the rationale come out complete without overrunning.
"Cursor is the AI code editor for professional developers who already work in an AI tool and want one that understands their whole codebase, not just the current line. Against Copilot it wins on agent multi-file edits and context; against the field it wins on product velocity and a voice developers trust. My read: this market is won at the editor and the usage data behind it, not the model."
The 60-second readiness statement
Rehearse the statement above until it's automatic. If you can state Cursor's positioning and your own market POV in one tight minute, you've internalized the role; if it comes out as a feature list, you haven't.
Go only when all five assets exist and every rubric dimension is a self-scored 4 or higher, scored against real artifacts. If a dimension is below 4, run the matching drill and re-score before you stop. Walking into a fast, founder-led loop under-rehearsed is the avoidable failure and a Cursor panel of engineers will find the gap you didn't close.
Takeaway. Confirm five assets exist, drill each dimension below a 4, re-time both rounds and only go when you can state Cursor's positioning and your POV in under 60 seconds.