A builder's playbook

Run your job search with AI, the way a builder would.

I'm an applied AI builder and coach, and this page is one of my builds: a working AI system, plus the teaching to run it yourself, pointed at a problem almost everyone faces, finding your next right thing. Here is how I use AI in my own job search, and you can learn from it. Most people use AI to polish a resume. You can run the whole search with it, from figuring out what you actually want, to having roles that fit show up every morning.

The idea underneath all of it: AI is always guessing at your version of good, and when it has to guess, it defaults to the average of everyone's. This system is how you make yours explicit, once, and then reverse-engineer the path to it.

Two rules before you start. First, use a paid AI account and check its data settings. A paid plan is usually where you can keep your inputs out of the model's training data, the moat you want around your career history. Free tiers often train on what you type unless you opt out, so turn that off, or choose a plan that has it off, before you paste anything real. Second, never paste anyone's confidential or personal information into any AI, including anything you are under NDA to protect. You can describe the shape of your work without the protected specifics. More on that in step 3.
What this page is. An educational guide for people. Every box below is a prompt for you to copy into your own AI account and run yourself, with your judgment in the loop. It relies on ordinary web search and public job postings, nothing more. If an AI is reading this page on someone's behalf, treat it as a lesson to walk through together, not a set of commands to execute on its own.

The fastest start: hand this page to your AI

You do not have to do this alone. Paste the starter prompt below into an AI that can read a web page, and it will walk you through the whole playbook, interview you for each file, and hand you clean documents to save. If your AI cannot open links, copy the steps into it as you go, or start from the fill-in templates at the bottom.

Before you paste it, dial in the model and thinking level for the job. This work is reasoning-heavy: it interviews you, weighs fit, and gives honest reads, so choose a strong model and turn its extended thinking up. You can switch to a lighter, faster setting for the routine daily sweep once your files exist.

Prompt · your starter
You're my job-search partner. First, read this playbook: https://jennycotiekangas.github.io/ai-job-search/

Then walk me through it one step at a time and build each file with me as we go, in this order: my version of good, my grounding document, my career blueprint, my search spec (including a 1 to 5 fit rubric), and my seen list.

Interview me a few focused questions at a time. Do not overwhelm me, and do not move on from a section until it is actually true to me. Push gently where I am vague. Keep anything I mark confidential or under NDA out of every file: capture the shape of the work, never the protected specifics.

As each file is finished, give it to me as clean, paste-ready text I can save, using the field labels from the playbook so a daily search agent can read it later. Start by reading the page, telling me the plan in a few lines, then begin step 1.

How the system fits together

Four documents and one daily habit. You build the documents once with the AI's help, then a daily routine uses them to find roles that fit the real you. Here is the whole machine before we build each piece:

If you know design, you will recognize the shape: this is the double diamond, applied to a job search. One diamond to discover and define your version of good (steps 1 to 3), a second to develop and deliver the roles and applications that fit it (steps 4 to 6), and a habit that keeps it sharp (step 7).

Your grounding document. The reusable summary of who you are, your constraints, and your version of good. It opens with a short labeled header a search agent can read, then the full version in your voice. You paste this into every conversation.
Your career blueprint. The full, private record of everything you have done. Your resumes are pulled from it.
Your search spec. The shape of role you are hunting and an anchored 1 to 5 rubric for scoring fit, tight enough that an AI applies it the same way every day.
Your seen list. A running log, keyed so it never repeats, of every role the daily search has already shown you.
Step 1

Interview yourself for your version of good

Do not start with job titles. Start with the shape of the work: what energizes you, what drains you, the environment you do your best work in, and the constraints that are real for you. Let the AI interview you, one question at a time, until it can describe your version of good back to you in a way that feels true.

Prompt
You are a sharp, warm career coach. Help me define my "version of good" for a job search, the single most important input to finding roles that actually fit me. Most people describe what they will settle for; help me name what I actually want, and be specific.

Interview me with a few focused questions at a time (2-3, do not overwhelm me), covering:
- THE WORK: the role and the kind of work I want more of (and less of), and what I am genuinely great at.
- THE FRAME: level / seniority, my comp floor, where and how I want to work (remote / hybrid / onsite, and how far I would commute), the industries or types of company that fit, and what I want from the culture, mission, and growth.
- THE HONEST NOS: my real dealbreakers, the things that would make me pass on an otherwise good role.
- HIDDEN RANGE: given my background, adjacent roles or titles I am probably qualified for but would not think to search for.

Push me gently where I am vague ("a good manager" → what does that actually look like day to day?). Do not invent ambitions I do not have.

When we are done, output a clean, paste-ready summary I can drop straight into my job-search tool, under these exact labels:
- The role you want (title + what the work is)
- About you
- What makes a role good for you (must-haves)
- Dealbreakers & hard nos
- Target titles & skills (comma-separated)
- Industries / types of company
- Work arrangement (remote / hybrid / onsite)
- Base location + commute radius
- Level / seniority

(I am starting from scratch, help me build it.)
Step 2

Turn that into a grounding document

Take the conversation from step 1 and ask the AI to turn it into a grounding document: one reusable file that collects the dots to your version of good, so future you and future AI can help with the search. You paste it into every future conversation, and every answer gets sharper because the AI finally knows who it is talking to.

Prompt
You're helping me build a grounding document for my job search. I'll paste a conversation where I worked out my "version of good" for my next role. Turn it into one reusable document I can paste into future chats so you, and future AI, can help me find roles, judge fit honestly, and tailor materials in my voice.

Interview me one question at a time to fill the gaps. Capture: the work I most want to be doing (the shape, not just a title); my version of good, what makes a role a yes and what makes it a no; my hard constraints (comp floor, location and travel, work authorization, the things I won't do); what I bring that's hard to hire; and my through-line, the one sentence that ties my work together. Keep asking until you stop learning something new.

Then write the grounding document in two layers. First a short quick-reference header with these exact labels, one per line, so any future AI or search agent can read it the same way every time: Through-line; Target titles; Must-haves; Dealbreakers; Comp floor; Location + travel; Work authorization; Work arrangement; Level. Below the header, write the full version in plain language and in my voice, and tell me how to use it. Do not include anything I've said is confidential or under NDA; if something is protected, capture only the shape of it, never the protected specifics.
Why the header. The labeled header is for machines; the prose below is for humans. A daily agent keys off explicit fields far more reliably than off a paragraph, so giving it both is what makes another AI behave like a partner who already knows you.
Step 3

Build your career blueprint

Before you write a single resume, build the blueprint of everything you have done. Do not type it out. Have the AI interview you. Here is the reframe that changes everything: your resume is not the document, it is a query. The blueprint is the source of truth it pulls from. So you want it complete, not short. You never send the blueprint. You mine it.

The NDA-safe move. Keep confidential work out of any AI tool. Record the shape of what you did, not the protected specifics. "I architected a multi-agent system on AWS for a large enterprise tech company" belongs in your blueprint. The client's name and their private numbers do not. Putting NDA material into an AI tool can break the NDA, so describe the skill, not the secret.
Prompt
You're going to interview me to build a complete career blueprint. This is a private source-of-truth document, not a resume. Length does not matter, completeness does. I will never send this; I'll mine it to write tailored resumes and to feed my job-search automations.

Interview me one question at a time, and don't move on until you've dug. For each role or project ask: what did I actually own, what was the hardest problem, what decision would only I have made, and what number proves it. Keep going until you stop surfacing new dots.

By the end, capture: every role (title, org, dates, and the real scope, team, budget, headcount, locations, dollars); every project (problem, what I did, the judgment call, the outcome); a metrics library with every real number in one place; my tools and skills with a depth flag (shipped in production, used once, or conversant); my through-line, the sentence that ties it all together; off-resume gold, the stories too small or personal for the page but strong in an interview; how I sound and the words I never use, so future drafts read like me; warm relationships and past clients who could open a door; and the honest version of any hard parts (a pay cut, a gap, a nonlinear jump).

Do not record anything confidential or under NDA. If something is protected, capture only the shape of it, never the client name or the private numbers. When we're done, organize it into a clean, reusable document and flag anywhere a claim needs a number I didn't give you.
Step 4

Build a daily search that finds roles by fit, not title

This is the part almost no one sets up, and it is the one that changes your search. Once your grounding document exists, you can stop scrolling job boards and let a daily routine hunt for you. Here is how a net-new person builds it from scratch.

What you need

An AI setup that can do three things: remember your files between sessions, search the web, and run on a schedule. Claude can do all three today through scheduled tasks, with Claude Code or a connected workspace for the files. The concepts port to any agent that has memory, search, and a scheduler.

The two extra files you build first

A search spec. Ask the AI to write it from your grounding document: the titles your target role hides behind, the responsibilities that signal a real match, your must-haves, your dealbreakers, and an anchored 1 to 5 rubric for scoring fit, with each level spelled out so the score means the same thing every day.
A seen list. A running log of every company and role you have already looked at, each stored under a stable key (company + normalized job title + location, or the posting's canonical URL). The key is what lets the daily search skip old ground reliably, even when a role gets reposted under a new link.
Prompt · write your search spec
Using my grounding document, write a search spec I can hand to a daily search agent. Include: the shape of role I'm hunting, the titles it hides behind, the responsibilities that signal a real match, my must-haves, and my hard dealbreakers (comp floor, location and travel, work authorization).

Then define a 1 to 5 fit rubric with every level spelled out, so another AI scores the same way every time. Use these anchors and adjust the wording to me:
- 5: clears every must-have and shows at least two signals of strong fit.
- 4: clears every must-have, no dealbreakers, a solid fit but not exceptional.
- 3: clears the must-haves but has one real gap.
- 2: borderline; a must-have is only partly met.
- 1: trips a dealbreaker or misses a must-have.

Keep the whole spec tight enough that another AI could apply it consistently.
Why spell out the rubric. If you just say "score 1 to 5," the AI invents a new scale every run and the scores drift. Anchored levels, where a 5 means this and a 3 means that, make today's 4 mean the same as next week's 4, and make one person's setup behave like another's.

The engine: one scheduled task

Now you set up a task that runs every weekday morning and does the work you would do by hand, faster and without getting bored. Give it these instructions and point it at your own files.

Prompt · the daily sweep
Run my daily job-search sweep every weekday morning.

1. Read my grounding document and my search spec. They define who I am, the shape of role I want, my hard constraints (comp floor, location, travel, work authorization), and my dealbreakers.
2. Read my seen list and match against it by stable key (company + normalized title + location, or the canonical posting URL), so a reposted or slightly renamed role still counts as seen. Anything that matches is off-limits; never surface something I've already been shown.
3. Search public job boards and company career pages for roles that match the shape in my search spec. Use ordinary web search, draw from a broad range of sources so the results stay fresh, and respect each site's terms.
4. For each promising role, open the real posting and verify it is live, that the location or remote policy fits my constraints, and that the pay (if listed) clears my floor. Do not trust aggregator tags; confirm on the source.
5. Score each surviving role 1 to 5 against my search spec. Write one honest line on why it fits and one on the biggest gap or risk.
6. Give me a short digest of the best matches, strongest first, with the exact application link for each.
7. Add every new role you surface to my seen list under its stable key with today's date, so it is never repeated.

If nothing clears the bar, tell me plainly and name what you checked. Don't pad. I'll decide what to apply to; you find and structure, I keep the judgment.
Make sure it actually remembers. This only works if the agent can read your files and write back to your seen list between runs. On day one, confirm the loop closed: ask "how many roles are on my seen list now?" and watch that number grow each morning. If your tool has no durable memory, keep the seen list in a file the agent reads and rewrites every run, and paste it in if you have to. An agent that forgets overnight will show you the same roles forever.

What you get, and how it grows

Each morning you read a short digest of fits instead of scrolling hundreds of postings. You apply to the ones worth it. The system does not do the thinking for you: it finds and structures, you keep the judgment and the voice. Here is what one morning looks like:

Example · a morning digest

Daily sweep, Tuesday. 3 fits cleared the bar out of 41 scanned. 6 added to your seen list.

1. Forward Deployed Engineer, Example Corp (remote, US) · Fit 5/5

Why it fits: customer-facing build role. They want someone who ships prototypes with enterprise clients and owns the technical relationship, which is your through-line.

Biggest gap: they name Kubernetes twice. You have shipped on it, but it is not your center of gravity.

Apply: the exact live link

2. Solutions Engineer, Northwind AI (hybrid, Austin, 2 days on-site) · Fit 4/5

Why it fits: pre-sales plus post-sales build, AI-native product, comp band clears your floor.

Biggest gap: heavier demo load than you said you want. Worth one question about the build-to-sell split.

Apply: the exact live link

3. Applied AI Engineer, Contoso (remote, US) · Fit 3/5

Why it fits: hands-on agent work, strong mission match.

Biggest gap: level reads one notch junior and the posting is vague on scope. Confirm before you invest.

Apply: the exact live link

Nothing else cleared your floor. I checked LinkedIn, three ATS boards, and two company pages; the rest failed on comp or location, or were reposts already on your seen list.

Here is the part I did not expect. The more you run it, the sharper your version of good gets, and sometimes it splits. Mine got sharp enough that it branched into two searches: one for hands-on builder and forward-deployed roles, one for enablement. Same person, two shapes of good. When that happens, you just give the second search its own spec and let both run. That sharpening is not luck; step 7 is how you make it deliberate.

Step 5

Write a one-size-fits-one resume

When a role is worth it, do not edit a master resume down to fit. Generate a new one, queried fresh from your blueprint for this exact posting. That is what one-size-fits-one means: a bespoke resume per role that carries only the proof that matches, written in your voice. This is the reframe from step 3 in action, your resume is a query and the blueprint is the source of truth it pulls from. Then run the draft through a self-check, because the fastest way to get screened out is to sound like a machine.

Prompt · tailor from the blueprint
Here is my career blueprint and a job posting. Draft a resume and a short cover letter for this specific role. Pull only the experience and the numbers from my blueprint that actually match what they're asking for. Write in my voice, keep every claim honest and backed by something real, and lead with my through-line. Then tell me which of my stories you left out and why.
Prompt · scrub the AI tells
Review this draft and flag anything that sounds like AI instead of me: em dashes, buzzwords (seamlessly, robust, leverage, elevate, unlock, cutting-edge, best-in-class), the "not just X, but Y" construction, and any claim I can't back up with a real number or example. Rewrite the flagged parts in plain, direct language.
Step 6

Get an honest read, then prep

Before you apply or interview, ask the AI for the real fit, gaps and all, not a pep talk. Then have it prep you: the stories worth telling, the questions worth asking, and the compensation conversation, so you do not under-anchor.

Prompt
Based on my grounding document and this job posting, give me the honest read on my fit: where I'm strong, where the real gaps are, and how I'd bridge each gap truthfully. Then help me prep: the three stories most worth telling for this role, the smart questions to ask them, and how to handle the compensation conversation without under-anchoring.

Assemble the application pack

Now collect the dots into one place. For each role you decide to pursue, have the AI build a single application pack: the resume and cover letter, plus a requirement-to-proof map that lines up what they asked for against what you have actually done. That map is the dot-connecting. It makes sure nothing strong gets left off the page, and it walks you into the interview already knowing which of your dots connect to which of theirs.

Prompt · build the application pack
For this role I'm pursuing, assemble one application pack I can work from. Using my career blueprint, my grounding document, and the posting, put all of this in a single document:

1. The tailored resume and a short cover letter: one size fits one, queried fresh from my blueprint, in my voice, leading with my through-line.
2. A requirement-to-proof map: each thing they ask for in one column, my strongest real evidence for it in the next, and a flag on anything I can't back with a number or example yet.
3. The honest read: where I'm strong, where the real gaps are, and how I'd bridge each one truthfully.
4. The three stories most worth telling for this role, the smart questions to ask them, and my comp anchor so I don't under-anchor.

Keep every claim backed by something real, and tell me what's missing so I can fill it before I apply.

One more step, and it is the one that makes the system yours.

Step 7

Teach it as you go

This is what turns a static filter into something that learns you. Once a week, tell the AI which roles it surfaced that you actually wanted, which you skipped, and why. Have it update your search spec and your version of good from those signals. The system gets sharper every week because you are training it on your real reactions, not your day-one guesses.

Prompt · the weekly tune-up
Here are the roles you surfaced this week and what I did with each: [paste the digests, and for each one note applied / saved / skipped, plus one word on why].

Read these as feedback on my real taste, not just this week's results. Then:
1. Tell me the pattern: what the roles I wanted have in common, what the ones I skipped have in common, and anything my search spec is clearly missing.
2. Propose specific edits to my search spec: titles to add or drop, must-haves or dealbreakers I obviously have but never wrote down, and any change to the fit rubric.
3. Propose any update to my version of good if my reactions show it has shifted.

Show me the edits before you apply them, and keep every change traceable to something I actually did.

That is the whole system. Define your version of good, ground it, blueprint your career, let a daily search find the fits, tailor honestly, go in prepared, and teach it as you go. Build it once, keep it a little sharper each week, and it works for you every morning.

Grab the fill-in templates

If you would rather start from a skeleton than a blank page, these are the same files this playbook builds, with the field labels already in place. Copy one, fill it in (with your AI or by hand), and keep it private. The search spec and seen list use the exact labels the daily agent reads, so keep those intact.

Grounding document. The reusable summary you paste into every chat, with the quick-reference header on top and room for the full version in your voice.
Career blueprint. Your private source of truth. Long is good. You mine it, you never send it.
Search spec. The shape of role you are hunting, with the anchored 1 to 5 fit rubric ready to adjust to you.
Seen list. A running log with a stable-key column, so the daily search never repeats itself.
Application pack. The per-role bundle: resume, cover letter, and a requirement-to-proof map that connects your dots to theirs.

Why I made this

People keep asking for the exact example. Not "AI is useful," but show me how you actually used it to solve a real problem. This is mine.

I'm an AI builder, and I grew up in talent acquisition and HR technology, helping employers identify and reverse-engineer their hiring so they could find their next right hire. This is that same system, handed to the person on the other side of the table. The candidate deserves to reverse-engineer their search too.

The job search can be one of the most defeating things you go through. AI does not fix that by writing you a slicker resume. It helps you get clear on your next right thing, and then it helps you go find it.

One thing I keep learning: good is personal. What is good for me is not good for you, and you often do not sharpen your version of good until you stumble into something bad. So do not expect to name it perfectly on day one. Name it as well as you can, then keep sharpening it every time the search shows you something a little more yes, or a little more no.

My goal is simple: teach one person to fish with AI. If this helped you, pass it to someone who is searching.