← Back to writing

Automation Playbook

How I Run My Job Search in AI Era

✦7 min

I apply to a lot of jobs. For cold applications sent through the AI pipeline, about 12% turn into interviews. Referrals and reaching out to hiring managers directly do much better, and I cover both below. The cold volume only works because each application is aimed carefully. I run my search on two open-source Claude Code skills, ai-job-search and career-ops, which I merged into one pipeline.

One caveat before the details. I have had roles with a match score over 85, an ATS score over 90 and a resume that mirrored 90% of the job description, and still got rejected within minutes. If that happens to you, it is usually the company's process, not you. I once read a Reddit post about an engineering lead who couldn't find anyone to hire. HR told him very few applicants passed the bar, so he applied to his own opening. He was rejected in six minutes.

A lot of postings are also probably not real openings. No pipeline fixes either problem. What it does is keep me from spending effort on roles that never fit.

Where I search

Lenny's Jobs comes first. Lenny has already filtered the companies on it, so I don't have to work out whether a team is any good before I apply.

LinkedIn hiring posts and groups come second. The LinkedIn job list is full of ghost jobs and roles that have sat open for months. When a hiring manager or recruiter posts about a role themselves, or someone shares it in a group, someone is usually trying to fill it right now.

YC Work at a Startup is the third. Founders are easy to reach there, and salary and equity are on the listing.

Finding real openings on LinkedIn

  1. Search posts instead of jobs. Type "hiring" plus the role, such as hiring product manager, filter by Posts, and limit it to the past day or week.
  2. Look at who posted. A hiring manager or recruiter writing under their own name is the best sign. A company page resharing an old listing is a weaker one.
  3. Check the company's careers page to confirm the role is still open.
  4. Join groups for your field. Hiring posts there tend to have a real person behind them who answers.
  5. Reach out the same day. I don't pay for Premium, so I get 3 connection requests with a personal note each month, and I save them for the companies I want most. For every other post, I leave a two-line comment on why I fit and send a plain connection request to the person who posted. About half of these get a reply.
  6. Apply on the official site as well, so you are in their ATS.

My pipeline, step by step

1. Build a detailed profile. Follow the README in career-ops and write your profile as a .md file. I position myself as a growth product manager who builds agentic AI products. Then I put every project I have done into the profile, including the ones that don't fit that headline neatly. The AI matches the profile against each job description, so a fuller profile finds more roles. My B2B projects are in there too, for example, and they are why I get SaaS interviews. The profile also sets the rules: numbers stay exactly as they happened, and there is a list of things I have never done and must never claim.

2. Pull new jobs every day. The job-scraper skill checks Lenny's Jobs, then LinkedIn, and skips anything it has already seen. Staffing agencies and aggregators are removed before scoring, so only direct employers are left.

I only apply to jobs posted in the last 3 days, in order of freshness. Everyone says to apply within 24 hours, and there is a reason: many roles get more than 100 applicants on the first day. So the last 24 hours come first, then the last 3 days. If that still doesn't fill my 30 applications for the day, a week old is fine. Roles that need specific domain knowledge usually stay open longer, so they are worth applying to even after the first day.

3. Score each job. career-ops evaluates every posting on role fit, matching evidence, gaps, level and company. A job needs 80 or more to continue. If its domain has nothing to do with my experience, the score is capped at 79. Years of experience, salary and location never count against a role.

4. Tailor the resume and check ATS. Each resume is one page, keeps all four of my roles, and tells the same true facts from the angle of that job description. A keyword scan then compares the resume with the posting. I only apply at 90 or higher, and I only add a missing keyword if I have actually done that work.

5. Submit and track. Every application gets a row in Notion with its score, resume file, form answers and status.

Three MCP connections keep the tracker current. Email moves a row to rejected, interview or assessment when the message arrives. Calendar holds interview times, and automated applying pauses around video interviews. Notion stores the whole pipeline, so I can check it from my phone.

Interview prep

Build your own story bank. Write down your best work as short stories with the situation, what you did and the result, and keep them in one file. Both skills include interview prep, and it works much better when it can pull from your stories.

Make sure you can use your stories for every job. The same story can show growth, AI work or cross-team leadership, depending on what the role asks for. Before each interview, pick the ones that match the job description and practice telling them from that angle.

Practice case studies too. Many PM loops include one, and they are hard to improvise.

For product sense questions, I built my own practice app, HatchFellow. You answer one PM interview question a day and get AI-assisted feedback on it.

An example

A Senior Product Manager role at an online marketplace scored 82 on match. My seller-side growth and marketplace work lined up well. The tailored resume told the same projects through seller onboarding and conversion, and it scored 97.6% on ATS, so it went out that day and went into Notion.

Mistakes I made

I used to cut every bullet that didn't look directly relevant. The resumes got thin, so now I keep transferable results and reframe them.

I applied through staffing agencies. They used up my daily quota and nothing came of them, so they are filtered out now.

I trusted high scores in domains I had never worked in. Those got rejected anyway, which is where the 79 cap came from.

I logged in to a batch of application sites and let the automation fill them in later. Most of the sessions expired first. Now I sign in and submit one at a time.

Each of these became a written rule in the pipeline.

Other things that helped

LinkedIn SEO. I rewrote my headline and About section around the keywords recruiters search for. I never paid for Premium, and my profile views still rose sharply, along with recruiter messages.

Referrals. They still work when the resume is strong. If someone offers to refer you, say yes. I hope to do the same for someone else later.

I'm Taiwanese, and the Taiwanese community has several platforms where people help each other with referrals, such as 歹丸郎 ReferHero and Nex Work. If you're not Taiwanese, look for the same kind of resource in your own immigrant community. For a general option, try refer.me.

Mentors. DaDaFly and ADPList let you book a free chat with a mentor.

Pay. The same kind of role often pays 70 to 80% of what it did two or three years ago. Some large companies I interviewed with had no room to negotiate, finance companies especially.

AI evals. Being able to show how you measure an AI feature matters a lot right now. Have your own AI projects ready to show.

let's build something. ✦

Happy to talk about products, AI and the calls behind them.

Say hello