How to Run Your Entire Recruitment Pipeline From Claude or ChatGPT (Without Wiring a Single Tool)
Most agencies pay for 8 to 20 recruiting tools and spend their week moving data between them. Here's how to run the whole pipeline from one AI agent instead, using the tools you already pay for.
Skip the read. Chat with this article.
Most recruitment agencies we talk to pay for somewhere between 8 and 20 recruiting tools.
A sourcing database. Two or three enrichment providers. A sequencer. LinkedIn automation. A job scraper. The ATS. And then n8n, Make or Zapier holding the whole thing together with tape.
And the honest truth is, the tools are not the problem. Every single one of them is good at its job. The problem is everything that happens between them: exporting a list here, uploading it there, checking which contacts are already in the CRM, re-running enrichment because the first provider missed 40% of the emails.
That in-between work is where your recruiters' week goes. So we built something to remove it.
Disclosure up front: hyreflow is our own product. This post explains what it does, how a real request runs through it, and who it is (and is not) for.
The Problem: Your Stack Grew, Your Leverage Didn't
Every new tool promised to save time. Most of them did, inside their own tab.
But a recruiter's job does not live inside one tab. A single BD play touches a signal source, a company database, a people search, an email finder, a verifier, a sequencer and the CRM. That's seven handoffs, and every handoff is a CSV, a copy-paste or a fragile automation.
We covered this in detail in the 5 levels of recruitment automation maturity. Most agencies are stuck at Level 2: lots of tools, very few connections between them.
The usual fixes all have a catch:
- Stitching it together in n8n, Make or Zapier works until a vendor changes their API. Then it breaks at night, and nobody on the team can debug it.
- Single-vendor AI add-ons only work inside that one vendor's tools and roadmap.
- Closed-box AI recruiting platforms ask you to rip out your stack, and you can't see what the AI actually did.
- Hiring more researchers is a linear cost. It doesn't compound and it doesn't run at night.
What a Recruitment Harness Actually Is
Think of it as brain and hands.
The brain is the AI agent you already use: Claude, ChatGPT, Claude Code or Codex. It's great at planning, reasoning about a job description and writing outreach. What it can't do on its own is log into your ATS, run an enrichment waterfall or push a list into Lemlist.
The hands are the harness. hyreflow gives the agent authenticated access to 40+ recruiting and GTM tools through one connection. Every tool comes as a pair:
- An adapter that actually calls the vendor's API, built from that vendor's real documentation.
- A written playbook the agent reads before it acts. It covers auth, guardrails, pagination, costs and, most importantly, when to stop and ask you.
That second part is what makes it work in practice. The playbooks carry the lessons from 100+ client projects, so the agent doesn't have to re-discover that one provider pulls the entire database if you nest a filter wrong.
“Your recruiters shouldn't be the integration layer between twelve tools. Let the agent do the plumbing and keep the humans on the phone.”
One Request, Start to Finish
Here's what a real run looks like. A recruiter opens Claude and types:
"Here's the JD for a Senior Account Executive, SaaS, London. Check our CRM first, then find me 20 more strong candidates and get their contact details."
This is what happens behind that one message.
Step 1: Parse the job description
The agent reads the JD and pulls out the searchable parts: titles and their variations, seniority, location, must-have experience and the competitor companies worth targeting. It shows you the filters before it searches, so you can correct anything that's off.
Step 2: Check your own database first (crm-matching)
Before buying a single new contact, the agent searches the candidates you already have. The crm-matching recipe uses your CRM's own skill tags and custom fields to prefilter, then reads notes, past interviews and applications to score each person against the role.
This step alone surfaces people your team forgot they had.
Step 3: Source net-new candidates (jd-to-shortlist)
Then it goes to the market. The jd-to-shortlist recipe runs a general search plus a competitor lane, where people already selling in the same category get boosted to the top. It also runs a promotion-ready lane: people one level below the vacancy whose work history has outgrown their title.
Sourcing runs as a waterfall across several people databases. hyreflow handles the provider order, deduplication and paging, so you get one clean list instead of four overlapping exports.
Step 4: Enrich only the people worth contacting
The agent qualifies the longlist against the JD first and only enriches the survivors. Contact enrichment is also a waterfall: it tries the cheapest provider first, moves on when one misses, and only charges when it returns a usable result. Soft misses cost nothing.
Step 5: You approve, then it pushes
You get a ranked shortlist with the reasoning for each person. Nothing has been sent and nothing has been written to the ATS yet.
You approve the ones you want, and only then does the agent create the records in your CRM or drop them into a Lemlist or Instantly sequence.
That's the rule across the whole product: the agent acts, the human approves. It never spends credits or writes to a system of record without an explicit yes from you.
Beyond Sourcing: Recipes for BD
The same harness runs your business development. Every recipe below came out of real agency work, and most of them map directly to the plays in our 7 BD outreach plays breakdown:
| Recipe | Trigger | What the agent does |
|---|---|---|
| candidate-led-bd-campaign | You have a strong, placeable candidate | Finds companies hiring for that profile, identifies the hiring managers, drafts an anonymised teaser and a full sequence (the MPC Shot) |
| cv-to-jobs | A CV comes in | Matches the candidate to live openings and finds the hiring manager for each |
| funding-to-bd | A company in your niche raises a round | Qualifies it against your ICP, finds the decision-makers, writes a first line off the funding news |
| layoff-signal-to-poach | A company announces layoffs | Finds the displaced talent worth calling, plus a BD angle with the teams that remain |
| tradeshow-to-bd | An event in your niche is coming up | Maps exhibitors and speakers, qualifies them and drafts a soft pre-conference intro |
| linkedin-sparse-sourcing | The role barely exists on LinkedIn (nurses, trades, chefs, mechanics) | Finds local employers, mines their team pages and finds the people there |
| build-tam | You need a target account list | Builds and qualifies a list of companies that fit your ICP |
You don't have to remember recipe names. You describe what you want in plain English and the agent picks the right recipe.
Your Stack Stays
This is the part agency owners usually ask about first, so let's be clear.
hyreflow is not a CRM, not a sourcing database and not a sequencer. It calls the ones you already run. It works with Bullhorn, Loxo, RecruitCRM, JobAdder, Vincere and most other recruitment CRMs, and with Lemlist, Instantly, Smartlead and HeyReach on the outreach side. You can see the full list on the tools page.
On cost, you choose how to run it:
- Managed credits: one login, one bill, no separate vendor contracts.
- Bring your own keys (BYOK): connect the API keys you already pay for. Those calls cost zero hyreflow credits.
- A mix of both: your own key for the tools you're committed to, managed credits for the rest.
Every result shows what it cost, so there are no surprises at the end of the month.
Who It's For (and Who It's Not For)
hyreflow is built for boutique recruitment agencies with roughly 3 to 25 seats that already run several recruiting tools and want more output without hiring more researchers. Ideally there's at least one person on the team who's happy to install things and try them out.
It's probably not the right fit if:
- You're an in-house TA team at a large company. The recipes are designed around agency BD and delivery.
- You want a classic no-code dashboard. The agent is the interface here, and the dashboard is for oversight: runs, connected tools and credit spend.
- You don't want AI anywhere near your pipeline. That's a fair position, and we're not going to argue you out of it.
How to Get Started
- Create a free account at hyreflow.ai. The free plan comes with credits for your first searches.
- Connect it to your assistant. One install connects it to Claude, ChatGPT, Claude Code or Codex. No token copy-pasting.
- Ask for an outcome, not a task. Start with something you'd normally give a researcher: "Find me 20 Heads of Finance at Series B fintechs in London who joined in the last 6 months."
- Connect your CRM and sequencer once you trust the output, so the agent can check your own database first and push approved results straight in.
If you want the background first, start with our guides to Claude Code for recruitment agencies and the Claude connectors worth setting up. hyreflow covers the recruiting-specific tools that don't have a native connector yet.
The Bottom Line
The tools are already good. What's missing is the layer that lets one agent use all of them, safely, with a human signing off on every step that costs money or touches your data.
That's what we built hyreflow to be. It came out of 100+ client projects, and every recipe in it is something we've run for real agencies first.
Try hyreflow for free and run your first search inside the assistant you already use.
Frequently Asked Questions
Related Articles
Written by

Niklas Huetzen
CEO & Co-Founder
Niklas leads Automindz Solutions, helping recruitment agencies across the globe build AI-powered pipeline systems that deliver warm meetings on autopilot.
Connect on LinkedIn →Free Resources
Want more like this?
Our Resource Hub is packed with free guides, templates, and tools to help you build AI-powered recruitment pipelines.


