Last recruiting season I hit the point every student hits. I'd spent a whole evening tailoring one application, typing my GPA into my fourth Workday portal that week, and I was pretty sure nobody was ever going to read it.
So I started tracking which applications actually got responses. The pattern was boring and obvious once I saw it:
Responses mostly come down to volume, plus applying early.
The roles where I heard back were almost always ones I'd hit within a few hours of the posting going live. The ones I'd polished for days and sent on day four? Silence. By then a posting has hundreds of applicants and a recruiter is skimming the first batch. Applying early gets you read, and applying a lot means some of those reads turn into calls.
You can't do both by hand. So I built something that does.
The result first
The first real test was on myself: about 1,600 applications in 4 days, which turned into 10 interviews.
That's not a magic conversion rate. It's the numbers game working the way it's supposed to, without me burning every evening on forms.
Then friends started asking me to run it for them. Then their friends did. At some point a script on my laptop has to become a real product, so I turned it into Applyinator.
Here's how it works under the hood.
The architecture: four -inators
I split the system into four pieces, partly because it made the code cleaner and partly because naming things "-inator" kept me sane at 2am.
1. Profile-inator: set it up once
Everything starts from a profile: target roles, cities, grad date, your resume, and your real answers to the questions applications keep asking (school, GPA, links, start date, and so on).
The rule is that the system only ever uses answers you gave it. It doesn't invent anything.
2. Watch-inator: polling 15,000+ boards every 1-3 minutes
This is where the timing advantage comes from. Instead of scraping job aggregators that repost listings hours or days later, the Watch-inator goes straight to the source: company job boards on ATS platforms like Greenhouse, Lever, Ashby, Workday and SmartRecruiters.
It polls 15,000+ company boards every 1-3 minutes. When a new posting shows up, it's normalized (title, location, team, posting time) and run against your profile to see whether it's a match.
Two things mattered more than I expected here:
- Polling cadence vs. politeness. Checking 15,000 boards that often means you have to be careful about how you schedule requests, so you're fast without hammering anyone.
- Dedupe. The same role often shows up under slightly different titles, gets reposted, or appears across locations. Every posting gets a stable identity, so it never applies to the same job twice.
3. Apply-inator: a real browser, start to finish
When a posting matches, the Apply-inator opens a real browser and goes through the actual application the way you would: it uploads your resume, fills in fields from your profile, handles multi-page forms and submits.
I went with a real browser instead of trying to POST forms directly because ATS forms are wildly inconsistent. Every company customizes questions, field order and validation. A browser that reads the page and fills what's there holds up much better than per-ATS hacks.
Because the Watch-inator finds roles minutes after they go live, most applications go out within minutes of the posting. There's no cap on how many it sends overall. The numbers game only works if the volume is real.
4. Ask-you-inator: the never-guess rule
This is the part I'm proudest of, and the part I'd tell anyone building agents to copy.
Applyinator never answers legal, visa or work-authorization questions itself. No guessing, no "most likely yes", no letting a model improvise. A wrong answer to "Will you now or in the future require sponsorship?" is far worse than a missed application.
When a form asks something that isn't in your profile, or anything in that legal/visa/work-auth category, the application is held, and the question goes to one central inbox. You answer it once. That answer is saved and reused everywhere it comes up after that, so you're not answering "Are you authorized to work in the US?" 300 times.
From a systems point of view, a held application is just a pending state with a structured question attached. Once you answer, it resumes where it left off.
The MCP server: "what's waiting on me?"
Since I already live in Claude most of the day, I exposed Applyinator as a remote MCP server. You can connect it to Claude, Claude Code or ChatGPT and just ask:
what's waiting on me?
You get back the held applications and the exact question each one needs. Answer in chat and they continue. Most days that's the only way I touch it.
Designing that tool interface pushed the never-guess rule all the way down: the MCP side returns a pending question for you to answer, and the model on the other end never gets to fill in a legal answer for you either.
Lessons learned
1. Timing beats polish. I spent years believing a better cover letter was the lever. For most roles, showing up in the first batch matters more.
2. Your agent's "never" list is the product. Making it apply was the easy part. Deciding what it must never do, and making those cases cheap for a human to clear (one inbox, answer once), took the most work and is what makes people trust it.
3. Real browsers are worth the cost. They're slower and heavier than raw requests, but they survive the endless variety of real application forms.
4. Dedupe is a first-class feature. Without it, volume turns into spam: the same job twice, reposts, duplicates across locations. Getting posting identity right is what keeps high volume clean.
5. Build for yourself first. I didn't do market research. I had a problem, built the fix, and people around me wanted the same fix. That's the most honest validation I know.
Try it
Applyinator is live as of this week. It's built mainly for students and new grads going after tech, quant and corporate roles. Pricing is flat: $10/week, $30/month or $100 lifetime, with unlimited applications.
If you're in the middle of recruiting season and tired of retyping the same answers, take a look at applyinator.com. If you're building agents, I'd love to hear how you handle the "this needs a human" problem. Drop it in the comments.
Top comments (0)