Hundreds of jobs a day, almost all wrong

Hundreds of remote jobs appear every day, and almost all of them are wrong for a given team. Reading them is a full-time job.

Keyword filters miss the point: "no agencies", "US citizens only", "a junior role" and stale companies all look like matches.

We wanted agents that read every post properly, judge it the way an experienced business developer would, and explain each decision.

A collector and three agents

Job Hunter finds and qualifies remote jobs for our own companies, BinaryBits and TheDataHQ, and it is used every day. Code collects and checks; AI agents decide.

The product screen, shown with test data. Real screens use the same layout.
📥
A collector
Reads free job sources every few hours and keeps only jobs it hasn't seen.
🗂️
Sources that aren't fixed
An AI adds new job sources, tests them, tunes their search words and retires the ones that bring nothing. People can suggest sources too.
🔎
A scout agent
Reads each whole post and decides whether it's a real chance, for which business, with which profile, and why.
✍️
A writer agent
Researches the company and finds the real person who decides, with an address that is written somewhere, never guessed.
✅
A reviewer agent
Checks every claim against the sources before anything goes out.
🧾
Everything logged
Every job kept or skipped, with its reason.

From a new post to a checked email

AI agentAutomated stepPersonData or serviceJob sourcesa list the AI managesCollectorkeeps new jobs onlyScoutjudges each whole postResearcher & writerfinds the real deciderReviewerchecks every claimEmail Managerpaced sendingSource brainadds, tests, retires sourcesDecision logkept or skipped, with reasontuningnew postsnew jobsloggedkept jobsdraft and sourceschecked email
Code collects and checks, AI agents decide, and every decision is logged with its reason.
1
Collect new jobs
The collector reads the job sources every few hours and keeps only jobs it hasn't seen before.
2
The scout judges each one
The scout reads the whole post, not just keywords, and decides whether it is a real chance, for which business and with which profile. Every decision gets a written reason.
3
The writer researches and writes
For kept jobs, the writer researches the company and finds the real person who decides, using only addresses that are written somewhere or verified.
4
The reviewer checks
A reviewer agent checks every claim against the sources before anything goes out.
5
It's handed to the Email Manager
Checked emails go to the Email Manager, which sends them one at a time at a human pace.

Agents decide, code collects

There are no keyword rules. Code collects and checks; AI agents read, judge and explain. The agents run on Anthropic's Claude, through Claude Code skills.

WhoWhat it does
Collector (automated)Reads the job sources every few hours and keeps only jobs it hasn't seen.
Source brain (AI)Adds new job sources, tests them, tunes their search words and retires the ones that bring nothing.
Scout (AI)Reads each whole post and decides if it's a real chance, for which business, with which profile, and why.
Researcher and writer (AI)Researches the company, finds the real person who decides, and writes the email.
Reviewer (AI)Checks every claim against the sources before anything goes out.
Every decision explained

Every job, kept or skipped, has a written reason, so a person can check the judgement.

Quality

A reviewer agent checks every claim against the sources before anything goes out.

No guessing

Only addresses that are written somewhere, or verified, are used.

People stay in control

Kept jobs go to the Email Manager, which only sends reviewed emails at a human pace.

Technologies used

Public job sources that are free to read, our own Python code, and AI agents built on Anthropic's Claude.

Python Claude agents Public job APIs and RSS feeds

Challenges and how we solved them

❌ Challenge
A source returning every category, not just tech.
✓ What we did
The scout noticed it and the source was tightened.
Result: Less noise.
❌ Challenge
People seeking work posting in hiring threads.
✓ What we did
The scout reads the whole post, so it isn't fooled by keywords.
Result: Correct skips.
❌ Challenge
Guessed email addresses would bounce.
✓ What we did
Only addresses written somewhere, or verified, are used.
Result: No guessing.

120 jobs judged in about four minutes

10
Job sources read automatically
~4 min
To judge 120 jobs on a real run
Every job
Kept or skipped with a written reason

On a real run, the scout judged 120 jobs in about four minutes and kept only the few that fit, each with its reason. Business results will be added here once they are measured.

Last updated: November 2026

What this means for you

The same agents can read and judge almost any stream of opportunities.

  • Lead finding and qualification. Agents that read every lead properly and keep only the ones that fit, with the reason.
  • AI screening of opportunities. Tenders, partnerships or projects judged against your own criteria.
  • Research agents for sales or recruiting. Company research and the real decision maker, with every claim checked.

For a client we would point the same approach at your own sources and criteria. For client systems we build on PostgreSQL, MySQL, MongoDB or Redis, and cloud data warehouses where needed, chosen for the project's scale and team. See our AI agent development services.

Frequently asked questions

How is this different from job alerts?

Alerts match keywords. The scout reads the whole post and judges fit, the way a person would, with a reason.

Which sources?

Public job boards, APIs and feeds that are free to read. The list changes as the AI learns which sources work.

Does it scrape sites that forbid it?

No. Only sources that allow reading, and nothing behind a login.