Working in AILesson 36 of 85

Contents

Working in AI

Lesson 36 of 85. Sign in to keep your place.

Module 4 · Getting seen: the routes that put a person in front of your work

Writing to strangers, and what to say

8 min read

One message that gets a reply, and six that do notWhat the four sentences doProve you read their actual work, by naming itShow your own work in one clause, with a linkAsk a question only somebody who has tried theproblem could askRemove the pressure of a requestWhat kills a message“I hope this finds you well”A CV nobody asked for“Any opportunities in your team?” as the firstcontactFlattery with no specificsFour hundred words about your journeyAsking for a call before giving anythingSpecific messages to the person who does the work run somewhere between one in five and one in three.Templated ones run near zero, and everybody can tell the difference instantly.
Specific messages to the person who does the work run somewhere between one in five and one in three. Templated ones run near zero, and everybody can tell the difference instantly.

Figure

Why this works when applying does not

An application joins a pile. A message arrives in an inbox. The person doing the job is rarely contacted, has the problem you would be solving, and, crucially, a hiring manager’s internal recommendation moves through a company faster than any application ever will.

This is the highest-yield and worst-feeling activity in a job search. Most people do not do it, which is precisely why it works.

Who to write to

The person doing the work: a data engineer, an analyst, the engineering manager. Not the recruiter, who receives hundreds of these. Not the chief executive, unless the company has nine people.

Find them by looking at who speaks about the company’s technical work: conference talks, blog posts, GitHub contributions, LinkedIn posts, a podcast, an answer on a forum.

The four-sentence message

Text
Hi Ananya — I read your write-up on moving the reporting
stack off nightly CSVs. I built something much smaller with
the same problem: a delay tracker for a transport
cooperative, where three months of GPS timestamps were in
local time and eight in UTC with nothing labelling which
(https://…).

One question: when you backfilled, did you reprocess the
old rows or keep both clocks and correct at read time? I
went with reprocessing and I am not sure it was right.

No ask — I am moving into data engineering and reading
people who have done it.

What that message does: proves you read her actual work, shows your own work in one clause, asks a question only somebody who has tried the problem could ask, and removes the pressure of a request.

Reply rates on messages like this run somewhere in the region of one in five to one in three. Templated messages run near zero, and everybody can tell the difference instantly.

What kills a message

  • “I hope this email finds you well.”
  • A CV attached that nobody asked for.
  • “Any opportunities in your team?” as the first contact.
  • Flattery with no specifics: “I admire your work” without naming any of it.
  • Four hundred words about your journey.
  • Asking for a call before you have given anything.

Volume

Fifteen to twenty-five well-targeted messages a month is the right scale. Not two hundred: at that volume you cannot be specific, and non-specific is the same as not sending. Two or three real conversations a month, sustained for six months, is a network.

The channels, and their real constraints

LinkedIn. A connection request carries a note with a hard 300-character limit on the free tier, which is roughly the four sentences above compressed. That constraint is useful discipline. Premium raises message limits and does not raise reply rates; the message does that.

Email. Company address formats are guessable and usually consistent: first.last@, first@, finitial.last@. Send one; a bounce costs nothing.

Communities. Discord and Slack groups for your language, city or tool; local meetup groups; the forum where people actually answer questions. Here the sequence is reversed: be useful in public for three months first, then a message is from someone they recognise.

GitHub. If you have fixed a bug or improved documentation in a project they maintain, you already have a conversation.

The follow-up rule

Once, after seven to ten days, adding something new: not “just bumping this”. If there is still no reply, stop. Silence is usually workload, not judgement, and a third message converts neutral into negative.

The slow version that compounds

Answer questions in public for three months. In the community for your city or your tool, help people with the thing you learned last month. You are qualified to help anyone one step behind you, which is a much larger group than you think.

What this produces is not followers. It is fifteen to thirty people who have watched you be useful, and who will forward you a job when one appears, which is what “network” actually means, stripped of its unpleasant connotations.

The free path

LinkedIn free, community servers, and a guess at an email format cost nothing. Hunter and similar services have free tiers of a handful of lookups a month. The paid products (LinkedIn Premium, Sales Navigator, Apollo) are built for salespeople sending thousands of messages, which is the opposite of this method.

Where this does not work

In markets and employers where hiring is centralised and procedural (large IT services firms with campus intakes, public sector recruitment, graduate schemes) the process is the process, and a direct message will be politely redirected to the portal. Match the method to the employer category. Outreach is for startups, mid-size companies, agencies and remote roles; the application form is for institutions.

Every lesson here is open. Your progress is recorded once you pass the AI at Work exam.

Check your understanding

Meera sends 40 messages a week to people at companies she wants to join. Each opens by praising the company, summarises her background, and asks whether they have openings. She gets almost no replies. What is the correct diagnosis?

Pick the one you would defend. Your answer stays private.

Next lesson · 9 minPublishing, and the inbound it creates

Stuck on something?Ask Rukmini, your AI tutor

How to cite: Addaly (2026) ‘Writing to strangers, and what to say’, in Working in AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-careers/writing-to-strangers (Accessed: 13 October 2026).