The Written Take-Home
Research an account and write outreach that an engineer would actually answer
Account research that shows up in the first line
After this you can gather signals that make personalization unfakeable.
The grader can tell in one sentence whether you researched the account or guessed at it. The take-home is won before you write the email, in the half hour you spend reading what the company has already told the world about how it builds software.
Personalization that lands is specific to their engineering reality, not a stitched-in name. An engineer reading your email is asking one question: did this person bother to understand my team or am I row 4,000 in a sequence. Your research has to answer that in the hook.
Where the real signals livemine these before writing a word
Open roles leak the stack, the team size and the pain.
“Hiring 6 backend engineers” plus a monorepo migration is a story about velocity strain.
Public repos, language mix and a blog post on their CI or AI tooling.
An engineer who wrote that post is a warm, named persona.
A fresh round usually means an aggressive hiring plan.
More engineers shipping faster is exactly Cursor's wedge.
Bottom-up means devs may already run Cursor before you call.
Check job posts, tweets and Stack mentions; it changes the whole angle.
That last card is the Cursor-specific twist. This is a bottom-up product, so the question is rarely do you use an AI editor - it's whether the team has standardized. If you find scattered individual adoption, your reason to reach out becomes governance and team rollout, not a first introduction.
Find the trigger that justifies reaching out now
A good email has a reason it arrived this week and not last quarter. That reason is a trigger event and naming it credibly is most of your personalization.
- A funding round or a hiring spike on the engineering team.
- A public migration - new language, monorepo move, platform rebuild.
- A new VP Eng or CTO who tends to reset tooling in their first 90 days.
- A stated AI mandate or an exec talking publicly about AI-assisted development.
Map the persona before you pick the angle
- Persona
- IC engineer
- What they care about
- Daily flow, less boilerplate, staying in the editor
- Your angle
- Concrete workflow win - Tab and codebase-aware edits in their stack
- Persona
- Eng manager
- What they care about
- Team velocity, onboarding speed, consistency across the team
- Your angle
- Standardizing the team and ramping new hires faster
- Persona
- VP Eng / CTO
- What they care about
- Org-wide impact, security, spend, measurable output
- Your angle
- Rollout, governance and the case for AI-native development
| Persona | What they care about | Your angle |
|---|---|---|
| IC engineer | Daily flow, less boilerplate, staying in the editor | Concrete workflow win - Tab and codebase-aware edits in their stack |
| Eng manager | Team velocity, onboarding speed, consistency across the team | Standardizing the team and ramping new hires faster |
| VP Eng / CTO | Org-wide impact, security, spend, measurable output | Rollout, governance and the case for AI-native development |
Same product, three different first lines. Sending the IC pitch to a CTO reads as a blast.
“Saw you're from Ohio, go Buckeyes” is not personalization - it's a name-drop with no relevance to why they should care about Cursor. Graders read it as a trick to look researched. Tie your hook to their engineering reality: a repo, a role they're hiring, a migration they announced.
Cite your source subtly so the recipient knows the work was real, without sounding like a stalker. “Noticed the six open backend roles and your move to a Go monorepo” shows the homework in seven words. Quote one specific thing, then move straight to relevance.
Takeaway. Research until you can write a first line that is true only for this account - a trigger event plus the right persona's pain. If the hook would work on any company, it isn't personalization.
Self-check
QAn engineering team has scattered individual Cursor users but no team-wide rollout. What's the strongest angle for outreach to their VP Eng?
Anatomy of a cold email that converts
After this you can write a short, personalized email with one clear CTA.
A cold email to an engineer survives on restraint. It does one job: earn a reply, not close a deal. Every extra sentence is a reason to archive it.
The structure is small on purpose. Hook them with something only true for their team, connect it to a single concrete value, then ask for almost nothing.
- 1Relevant hook. One line proving you researched them - a trigger event or a specific engineering detail.
- 2One-line value tied to their context. What Cursor does for a team like theirs, in plain words, not a feature dump.
- 3Single soft CTA. One ask, low-friction - interest, not a calendar demand.
Interactive diagram. Tab through its regions; each focused region shows its detail in the panel below.
Each part earns the next. The CTA is the gate every email passes through - get it wrong and the rest is wasted.
- Length
- Roughly 50–125 words. If it scrolls on a phone, it's too long.
- Asks
- Exactly one. A second ask halves your reply rate.
- Buzzwords
- Zero. “Overhaul your SDLC with AI” gets deleted by engineers.
- Subject line
- Specific and human, 3–6 words. Not “Quick question.”
The CTA: interest beats a meeting demand
Asking a stranger for thirty minutes cold is a big request and engineers protect their calendar fiercely. An interest-based CTA lowers the bar to a one-word reply and a yes converts into the meeting on the next touch.
- Weak CTA
- “Do you have 30 minutes Tuesday at 2?”
- Why it fails
- Big ask, presumes interest, easy to ignore
- Stronger CTA
- “Worth a quick look or not a fit right now?”
- Weak CTA
- “Let's hop on a call to discuss synergies.”
- Why it fails
- Vague and salesy, no reason to reply
- Stronger CTA
- “Want me to send a 2-minute Loom on the monorepo angle?”
| Weak CTA | Why it fails | Stronger CTA |
|---|---|---|
| “Do you have 30 minutes Tuesday at 2?” | Big ask, presumes interest, easy to ignore | “Worth a quick look or not a fit right now?” |
| “Let's hop on a call to discuss synergies.” | Vague and salesy, no reason to reply | “Want me to send a 2-minute Loom on the monorepo angle?” |
The soft CTA gives them an easy yes or an easy no - both beat silence.
Subject: your Go monorepo + 6 backend hires Hi Priya - Saw you're hiring six backend engineers and just moved to a Go monorepo. Ramping that many people on a large codebase is usually where velocity goes to die. Cursor is an AI editor that reads your whole repo for context, so new hires get accurate, codebase-aware edits and explanations from day one instead of week three. A few teams your size use it mostly to cut onboarding time. Worth a quick look or is now the wrong time? - Alex
The hook is true only for Priya's team. The value is one sentence in her language - onboarding on a large codebase - not a feature list. There's a single soft CTA, it's under 100 words and the subject line is so specific it reads like a colleague, not a vendor.
Resist the urge to prove the whole product in the email. Listing Tab, Agent, codebase context and enterprise security in one message signals desperation. Pick the one capability that matches their trigger and leave the rest for the reply.
Takeaway. Hook → one-line value → one soft CTA, under ~125 words. The email's only job is a reply, so ask for less and you'll get more.
Self-check
LinkedIn and multi-touch messaging
After this you can adapt your message across channels without copy-pasting.
LinkedIn is a hallway, not an inbox. The register is shorter, more conversational and far less tolerant of a pitch dropped into a connection request.
The cardinal error is treating LinkedIn as email with a smaller box. Copy-pasting your cold email into a DM reads as automation and engineers spot it instantly. Each channel earns its own version of the message.
Room for a hook, one value line and a soft CTA.
Slightly more formal; they may forward it internally.
Shorter, warmer, conversational.
Lead with relevance; let the pitch come after a reply.
Interactive diagram. Tab through its regions; each focused region shows its detail in the panel below.
Not a smaller box for the same words. Each channel earns its own version of the message.
Don't pitch in the connection request
The connection request earns the right to message, nothing more. A pitch crammed into that 300-character note is the fastest way to get ignored or marked spam. Keep it to genuine relevance and let the reply open the door.
- Touch
- Connection request
- Job
- Earn the connection on relevance
- Tone
- Brief, human, no ask
- Touch
- First message after connect
- Job
- Open a conversation, light value
- Tone
- Conversational, one question
- Touch
- Follow-up
- Job
- Add a new angle or proof point
- Tone
- Helpful, not nagging
| Touch | Job | Tone |
|---|---|---|
| Connection request | Earn the connection on relevance | Brief, human, no ask |
| First message after connect | Open a conversation, light value | Conversational, one question |
| Follow-up | Add a new angle or proof point | Helpful, not nagging |
Each touch does one job; none of them is “close.”
Every touch adds something new
If your second message just repeats the first louder, you've taught them to ignore you. A multi-touch effort across email, LinkedIn and a call should feel like one person approaching the problem from different sides, never the same message three times.
Technical buyers research you back before they reply. They'll open your LinkedIn profile and if it's empty, off-brand or contradicts your message, your credibility evaporates. Match the engineer's register - direct, accurate, low-hype - and make sure your profile says the same thing your message does.
Connection note: “Hi Priya - saw the Go monorepo move and your backend hiring push. I work with eng teams scaling fast on Cursor and would love to connect.” No pitch, just a real reason and a clean open.
Takeaway. Re-write per channel - LinkedIn is shorter and warmer than email. Earn the reply before you pitch and never copy-paste the same words across touches.
Self-check
QWhat's the right job for a LinkedIn connection request when prospecting a skeptical engineering buyer?
Designing a coherent sequence
After this you can build a multi-touch cadence with a logical arc.
A sequence is a short story told across touches, not the same message sent five times until they break. Each step changes the angle so a non-reply on touch one doesn't doom touch three.
The take-home sometimes asks for the full sequence and graders look for an arc. The classic angles move from their problem, to proof it's solvable, to a peer who solved it, to a graceful exit.
- 1Problem. Open on their trigger and the pain it implies - the personalized hook.
- 2Proof. Show the win is real: a concrete capability or a quantified result for a team like theirs.
- 3Peer. Reference a comparable company or engineer who solved the same problem with Cursor.
- 4Breakup. A short, honest “I'll stop reaching out - wrong time or wrong person?” Often the highest-reply touch.
Interactive diagram. Tab through its regions; each focused region shows its detail in the panel below.
A short story across touches, not the same message five times. The breakup is the gate that surfaces who's actually alive.
- Day
- Day 1
- Channel
- Angle
- Problem - personalized hook + soft CTA
- Day
- Day 2
- Channel
- Angle
- Connection request, relevance only
- Day
- Day 4
- Channel
- Call
- Angle
- Quick voicemail referencing the email
- Day
- Day 6
- Channel
- Angle
- Proof - one capability or result
- Day
- Day 9
- Channel
- Angle
- Peer - a comparable team's story
- Day
- Day 12
- Channel
- Angle
- Breakup - graceful exit
| Day | Channel | Angle |
|---|---|---|
| Day 1 | Problem - personalized hook + soft CTA | |
| Day 2 | Connection request, relevance only | |
| Day 4 | Call | Quick voicemail referencing the email |
| Day 6 | Proof - one capability or result | |
| Day 9 | Peer - a comparable team's story | |
| Day 12 | Breakup - graceful exit |
Roughly two weeks, mixed channels, each touch a different angle. Spacing is sensible, not daily.
The breakup message earns surprise replies because it removes pressure and triggers loss aversion. “Sounds like the timing's off - I'll close this out unless I'm wrong” often gets a “wait, actually…” It's also honest, which fits an engineer's expectations.
Tie cadence back to the activity math
An SDR's pipeline is arithmetic and a strong take-home rationale shows you think that way. The funnel runs from touches to conversations to qualified meetings and your sequence is the engine that feeds it.
- Sequences started / week
- ~50 prospects entered into a 6-touch cadence
- Positive reply rate
- ~5% reply with interest → ~2-3 conversations
- Conversation → SQL
- ~30-40% become qualified meetings for an AE
- Output
- Roughly the meetings that sustain a monthly SQL quota
Treat these as illustrative ratios, not Cursor's published numbers - the point is showing you reason from activity to output.
Close your rationale by saying you'd measure reply rate and positive-reply rate per step, kill the weakest touch and A/B the subject line. Showing you'd iterate on data, not just send and hope, separates a serious candidate from someone who wrote one good email.
Takeaway. Sequence = problem → proof → peer → breakup, mixed channels over ~2 weeks. Tie it to funnel math and say you'd measure reply rate per step and iterate.
Self-check
Selling an AI dev tool without hype
After this you can earn an engineer's trust through accuracy and restraint.
Engineers are the hardest audience for a marketing voice on earth. They detect exaggeration instantly and punish it by tuning you out, so your edge in this take-home is precision, not enthusiasm.
The whole point of the written exercise is whether you can make a complex AI tool simple and accurate. Hype reads as a tell that you don't understand the product. Restraint reads as someone an engineer could trust.
Reference real capabilities, correctly
Product literacy is the cheapest credibility you can buy and getting a feature name wrong is the cheapest way to lose it. Know what Cursor actually does before you write a word about it.
Predicts your next edit across multiple lines.
The everyday flow win an IC feels first.
Makes multi-file changes from a natural-language task.
Frame as workflow, not “it codes for you.”
Answers and edits grounded in the whole repo.
Why onboarding and large-codebase work get faster.
Frame value as workflow, not magic
- Say what changes in their day: less boilerplate, faster onboarding, fewer context-switches out of the editor.
- Talk team standardization for managers - one tool, consistent practice, faster ramp.
- Keep claims testable: “codebase-aware edits,” not “10x your engineers.” Engineers will check.
“AI that writes your code for you” and “10x developer productivity” are credibility-killers with this audience. They're vague, unverifiable and slightly insulting to people who write code for a living. Describe the workflow change precisely and let them infer the value.
Acknowledge competitors honestly
If a team already runs Copilot or evaluated Windsurf, pretending those don't exist makes you look naive. Honesty about the landscape is more persuasive than spin - name the comparison and point to where Cursor is genuinely differentiated, like deep codebase context and the agent workflow.
HYPE (delete on sight): “Cursor's revolutionary AI will 10x your team and automate your entire codebase overnight.” CREDIBLE (an engineer keeps reading): “Cursor reads your whole repo for context, so edits and explanations are grounded in your actual code - most teams feel it first on onboarding and big refactors.”
“You're already on Copilot, so this isn't a first AI tool - the difference teams notice is how much Cursor pulls from the whole codebase, which is where most of the onboarding and refactor time goes. Happy to show the specific spots if it's useful.”
Takeaway. Be precise about what Cursor does - Tab, Agent, codebase context - and frame value as a workflow change, not AI magic. With engineers, accuracy is the persuasion.
Self-check
QA prospect's team already uses GitHub Copilot. What's the most credible way to position Cursor in your outreach?
Polishing to Cursor's bar
After this you can self-edit the exercise to the standard that gets an offer.
The take-home is a writing sample disguised as an exercise. Whoever hires you is asking whether they'd trust you to be the brand's first impression to a technical buyer and a sloppy submission answers no.
Cursor's bar on authenticity and craft is unusually high because it sells to skeptical engineers. The polish pass is where you separate from candidates who wrote a decent draft and stopped.
Cut until every sentence earns its place
Tightness signals respect for the reader's time, which is exactly the quality the job needs. Read your draft and delete any sentence that doesn't move the prospect toward a reply.
- Length
- Is the email under ~125 words and skimmable on a phone?
- One ask
- Exactly one CTA and is it soft enough to get a one-word yes?
- Buzzwords
- Zero hype, zero vague claims an engineer would challenge.
- Typos
- Proofread aloud - a single typo in outreach is a credibility leak.
- Voice
- Does it read as authentically yours, with no AI tells?
Assume AI assistance is off the table for this exercise. Cursor restricts AI in its interviews and a generic, AI-flavored email is easy to spot and a hard fail. The work has to read as authentically yours - the messy, specific, human details are the point.
Add a brief rationale if asked
Many take-homes ask why you chose this account, persona and angle. A short, sharp rationale is where you prove the thinking behind the writing and it's often weighted as heavily as the email itself.
- 1Why this account. The trigger event and the signal that made it timely.
- 2Why this persona. Who owns the pain and can act and why you chose them over a peer.
- 3Why this angle. The single value you led with and why it fits their reality.
Read every message out loud once. Check the recipient's name and company spelling twice. Confirm there's one CTA per touch and that no two touches repeat. Re-open your LinkedIn profile and make sure it's consistent with the voice you just wrote.
If the prompt is silent on rationale, include three or four crisp lines anyway. Volunteering your reasoning shows the judgment Cursor wants in an SDR who'll soon help build the playbook - and it gives the grader a reason to picture you as an AE.
Takeaway. Treat the submission as a writing sample: cut every dead sentence, proofread until it's clean, keep it authentically yours and add a short why-this-account/persona/angle rationale.