Three thousand applications.

The ritual is familiar to anyone who has looked for work this decade. A role goes up on a careers page. Within hours it has hundreds of applicants; recruiters, being human, typically review something like the first hundred. To be one of them you must find the posting before the crowd does, create yet another account on yet another applicant-tracking system, upload a résumé and then re-type its entire contents into form fields, and produce a sincere answer to “Why us?” — then do all of it again, tomorrow, for the next posting, for months. Companies automated their side of hiring years ago; the screening software reads you before any person does. The applicant, meanwhile, is still expected to work by hand.

For one group the ritual carries a countdown as well. An international student in the United States gets a fixed window of work authorisation in which to land a job, a sponsor and a future — every week of silence is a week off the clock, and half the postings quietly do not sponsor at all, a fact discovered only after the application is in. Speed, volume and accuracy decide the outcome, and the process is hostile to all three. The asymmetry is the point: the modern job hunt is a machine on one side and a tired person at midnight on the other.

This transmission is about a founder who lived that arithmetic personally — three thousand-plus applications, filed by hand, on the international student’s clock — and then, with a co-founder who had lived it too, built the machine for the applicant’s side. Readers will know the shape this series documents by now: decades of judgment, encoded into a system. This is the seventh time we have filed the story, and this month the formula inverts — the operator is at the start of his career, and the judgment is measured not in decades but in repetitions. Three thousand of them.

The operator in question

Pulkit — first names are policy here; he can introduce himself properly — is a computer-science student turned founder out of Rose-Hulman, the Indiana engineering school, and the résumé he was mailing out three thousand times is worth pausing on, because its owner had already shipped more than most careers do. A language-learning app built for a Government of India programme, used by five hundred thousand people. An AI translation pipeline for an Indian ed-tech company, processing thousands of files a month. A dealflow system for a capital advisory, syncing fifty thousand emails a month through a queue he engineered. A medical-intake voice agent — speech in, reasoning, speech out, designed to HIPAA constraints — built and deployed alone, on an internship. That is the profile that was going unread past position one hundred.

So the founding story is not an epiphany; it is a tally. In their own words, on their Y Combinator page: “We built Tsenta after applying to over 3,000 jobs manually, so that nobody else would have to.” He founded the company in May 2025 with a co-founder from the same computer-science programme — another international student with his own stack of internships and his own pile of applications — and the pair took it into Y Combinator’s Summer 2026 batch. His one-line bio for the period is the best summary of the genre we have read: “Building the job platform I was too busy applying to jobs to make.”

Three thousand repetitions of anything is an education. Somewhere in that pile — the Workday accounts, the re-typed résumés, the sponsorship questions answered wrong once and never again — sits a complete, involuntary field study of the applicant-tracking gauntlet: which forms break, which questions repeat, what recruiters actually see, where the honest hours go to die. Nobody would fund that research. Two people simply lived it, and then did the thing this series exists to document: they encoded it.

Four stages, one agent

Tsenta describes its pipeline in four verbs — find, prep, apply, track — under a tagline of appropriate bluntness: “Four stages. One agent. Zero spreadsheets.” The find stage is the product’s physics. Tsenta watches fifty-thousand-plus company career pages directly — Workday, Greenhouse, Lever, Ashby, nineteen applicant-tracking systems in all — and the moment a role appears that fits the profile, it is matched within seconds, before the job boards have noticed it exists. Their line for it: the role finds you, and you apply first — inside the hundred applications a recruiter will actually read, instead of seven hundred deep via the boards.

The prep and apply stages are where the trust engineering lives, and the design choice runs through everything: transparency over magic. The résumé and cover letter are rewritten per role from the actual job description — “using only true facts from the résumé you uploaded” — and every change is shown before anything goes out: “never silently sent.” The agent then opens the real form, fills every field, answers the open-ended questions in the applicant’s voice, handles the work-authorisation questions correctly from a status set once, filters out companies that do not sponsor — and files a receipt afterward: the exact fields, the exact answers, the documents that went, the confirmation back from the ATS. There is even a standing AI-disclosure page. For a category with a spam-cannon reputation, the posture is pointed: an agent acting in your name should be auditable, line by line.

It lives wherever the applicant does: web, iOS and Android apps, a Chrome extension that fills any posting in a click, an iMessage bot — a match arrives as a text; reply yes and the receipt comes back — and, delightfully for readers of this series, an MCP server and CLI, so your own AI assistant can run the search for you. The economics are the applicant’s economics: the first twenty-five applications free with no card, then tiers priced per application actually submitted — “pay for the applications, not the tool” — with the full product at every tier, and tier copy honest to the point of bruising: for “the desperate, the laid-off, the OPT-clocked.” Forty-five thousand people now use it, and by the founders’ launch numbers more than seventy per cent of paid users have landed interviews.

Notice, for the seventh time in this series, the shape — with the variable flipped. In our last transmission the encoded judgment was two decades of infrastructure; here it is three thousand repetitions of a modern ordeal, systematised while the bruises were still fresh. The lesson the series keeps finding turns out not to care how old the operator is. Judgment is accumulated in reps, not years; pain, taken seriously enough, is a spec; and on the specific subject of the ATS gauntlet, a student with three thousand filings may simply know more than anyone else alive.

Notes for anyone automating on someone's behalf

Strip away the job hunt and the rules travel to anyone building — or using — an agent that acts in a person’s name:

Speed is a qualification. The first hundred get read; merit that arrives late is merit unread. Know the window your field actually operates on, and build for the window, not the queue.

Encode the pain while it is fresh. Three thousand applications is market research nobody would ever fund. Your worst recurring grind, documented honestly, is a product spec waiting to be read.

Automate the filing, never the claim. Only true facts, reformatted per audience. An agent that lies on your behalf is a time bomb with your name on it; one that formats the truth is leverage.

Show the work or lose the trust. Approval before send, receipts after. Any system acting as you must be auditable by you — “never silently sent” is the whole contract in three words.

Charge for output, not access. Pricing per application submitted points the tool at the user’s outcome. A subscription rewards the vendor for your logging in; a per-result price rewards it for your getting hired.

The next three thousand

What changes for the applicant is the part worth stating plainly. The evenings come back. The posting is found in seconds instead of days, and the application lands inside the window where a human will actually read it. The answers are right the first time, sponsorship dead-ends are filtered before they waste a filing, and the inbox stops being a graveyard of no-reply addresses and becomes — their word — a pipeline. The search stops being a second job performed at midnight and becomes something closer to what it always should have been: a decision about where to go next.

And there is something fitting in where this one sits in our record. This series has filed six stories about operators encoding twenty-year careers. The seventh is the youngest operator we have covered, and the point of him is that the mechanism is identical: live the problem at full intensity, take it more seriously than anyone paid to solve it, write the judgment down until a machine can run it. His co-founder’s bio says he is building “the job automation platform I wish I had at 19.” They are, give or take, still close enough to nineteen to remember exactly what it needed to be.

So if you — or your kid, or your best engineer’s partner, or anyone else on the wrong side of the wall — are staring down a season of Workday forms, the agent is at tsenta.com, and the first twenty-five applications are free, no card required. We will give the last words to the sentence that explains the whole company: “We built Tsenta after applying to over 3,000 jobs manually, so that nobody else would have to.” They did. Nobody else does. The next three thousand are the machine’s problem.

— END TRANSMISSION 05J