we built a claude agent who's ONLY job is to scrape job descriptions for leads...

a job posting is a company telling you, in writing, what it is about to spend money on.

here is how to build a single agent that reads those postings and turns them into qualified leads....

---

why a job description is a lead source:

a company that posts a role has budget approved and a problem it is hiring to solve.

the description itself is a briefing document:

• the role names the function that is under strain
• the responsibilities describe the exact problem they need solved
• the requirements list the tools and the tech stack they run
• the team section shows who the hire reports to, meaning your decision maker
• the seniority tells you how big the gap is

a company hiring three sdrs has a pipeline problem and the budget to fix it. that is a CLEANER signal than any firmographic a database will sell you.

the timing is the other half.

a role posted this week is a problem the company is actively spending to solve,

so you reach them while the budget is open rather than months after it closed.

there is also a second layer most people miss.

a role that stays open for two months is a problem they cannot hire their way out of,

which is a company that would rather hand the work to a service than keep failing to fill the seat.

---

what the agent extracts:

the agent reads each posting and returns a structured lead, not a link.

the fields it pulls per posting:

• company name and domain
• the role and its seniority
• the tools and platforms named in the requirements
• the problem the role exists to solve, inferred from the responsibilities
• the likely decision maker, meaning the role the hire reports to
• location, team size, and how long the role has been open

then it qualifies. it checks the company against your icp and drops the ones that miss, with a written reason on every call.

the inferred problem is the valuable part.

the agent reads a jd for a "demand gen manager to build our outbound motion"

and outputs that this company is starting outbound from scratch, which is your exact opening line.

the extracted tools let you segment later.

every company whose jd names a competitor's product is a warm switch conversation,

and every company hiring for a tool you complement is a NATURAL FIT

---

how to build it:

build it as a single purpose agent with one loop, using claude code or an agentic model over the api.

the steps it runs:

• take a source, meaning a linkedin jobs search url, an ats feed from greenhouse, lever, or ashby, or a job board query
• pull the list of live postings that match your target roles and locations
• open each posting and read the full description
• extract the fields above into a structured row
• infer the problem and identify the decision maker from the reporting line
• qualify the company against your icp, keep or drop with a reason
• find the decision maker's profile and email
• write the row to a file

give it your icp and target roles as the system prompt, so every decision it makes is measured against your definition rather than a generic one.

the ats feeds are the most reliable source, because greenhouse, lever, and ashby expose their postings as public json with no login,

so the agent reads structured data instead of fighting a page.

build the inference step carefully, since it is what separates this from a job scraper.

you are turning each one into a named problem and a named person to contact.

---

running it:

point it at a schedule, new postings appear daily, and a role posted this week is the freshest signal you can get.

the guardrails:

• cap each run with a budget so a loop cannot overspend
• dedupe against companies you have already contacted
• verify every email before it reaches a domain
• smoke test on 100 postings first, check the field accuracy and the cost per lead, then scale

route the output straight into outreach with the problem already attached, so the opening line writes itself from what the jd told you.

track which inferred problems convert best.

if "building outbound from scratch" books more calls than "scaling an existing team",

you weight the agent toward the roles that signal the first, and the whole system sharpens on the data it generates.

a company that just posted the role is deciding how to solve that problem RIGHT now, which is the moment you want to be in their inbox.

---

- din

we run cold outbound systems (linkedin + email) for B2B businesses making $10k/month or more...

worked with over 140 companies across 30 different industries (most get ROI within 60-90 days)

DM me and we'll talk

also read this article for more value:

https://x.com/DinScales26/status/2062960719247643068