Most recruiters I talk to use Claude the way they used ChatGPT a year ago: open a tab, paste a JD, ask for a Boolean string, copy the answer, close the tab. That’s what Claude looks like in the first three days, before you’ve found the real product.
Three months ago I started using Claude seriously. In that time, it has done the work of what would otherwise have been three hires: a research analyst, an ops coordinator, and a content manager.
From what I see across our own team, most of a recruiter’s week is repeatable mechanics: screening, Boolean strings, “personalized” outreach that’s 80% template, weekly client reports off ATS funnel data, salary benchmarks, company research, interview summaries, candidate emails, JDs. This is exactly the work AI agents do better and faster — not because they’re smarter, but because they don’t get tired, don’t context-switch, and run in parallel.
The recruiters who don’t get fluent with this stack in 2026 will, in 12 months, be competing against recruiters who have a five-agent AI team running inside one Claude account. Guess who closes more. The difference shows up in output, not in convenience. Half of what you do by hand today simply stops being your job.
- Stop using Claude as a search bar. Start using it as an office.The single biggest unlock: Claude’s desktop app has Projects. Inside each Project you save documents, instructions, and standards — then create separate chats *inside* it, each one acting as a different role. The Project is the office. The chats are the people who work in it.The non-obvious rule from our internal guide: one chat = one task type. Always return to the same chat for the same kind of work. Claude remembers everything from previous messages in that thread — it gets smarter at *your* version of the job over time.We run two Projects. Sourcing Team holds our tone of voice, ICP, outreach templates that have actually worked, and a list of phrases we’ve banned (“exciting opportunity,” “I came across your profile”); inside it, chats for Search Queries, Outreach, Candidate Communication, Candidate Research, and Analytics.Recruiting Team holds our candidate-summary format, weekly client-report template, and JD standards; inside it, chats for Job Description, Candidate Summary, Client Follow-Up, Candidate Communication, and Analytics.
When a recruiter opens Candidate Summary and types “write the summary for the call I just had with N,” Claude already knows our format — 150–200 words, no buzzwords, ends with a clear recommendation. No one re-explains the rules every time.
If you do nothing else from this article, do this. Take 30 minutes this week. Build one Project. Drop in three of your best outreach messages, your ICP, the stack of your top client, and two Boolean strings that have worked. Create two chats: Search Queries and Outreach. Tomorrow will feel different.
- Use Deep Research instead of asking research questions.When recruiters say they want to “research the market,” they type a question and get an answer pulled from maybe 10 sources. That’s a slightly better Wikipedia summary, not research.Claude has a feature called Deep Research. Click one button and it spends ten minutes analyzing 300–500 sources for one query.A huge portion of what we get asked at EvoTalents is talent market intelligence. A client is expanding into a new geography, or opening a role they’ve never hired for before, and they want to know: where do those people actually live, what do they cost, what taxes apply, is the market saturated, does it make sense to open a hub there. That used to be hours — sometimes days — of pulling from Glassdoor, levels.fyi, local job boards, Numbeo, salary forums, and country-specific tax calculators, then stitching it together. Now it’s one Deep Research run.
Recent example: a client wanted to know where to open a Senior DevOps hub — Poland vs. Portugal vs. Romania vs. Spain. I asked Deep Research for talent-pool size, gross-vs-net salary, employer-side taxes, market saturation, remote-vs-onsite norms, and 18-month trends per location, ending with a comparison table and a recommendation. Ten minutes later: a structured report with citations to 326 sources and a defensible answer. The client made the hub decision the same day instead of three weeks later. That used to be a research-analyst job.Other places I now run Deep Research instead of Googling for two hours:
-
- Salary benchmarks for new geographies or stacks
- Pre-BD-call competitive intelligence on a prospect
- Deep dives into emerging verticals (e.g. “how are companies hiring AI Research Engineers in the EU”)
- Quarterly talent market reports for clients
- Sourcing universe mapping (“top 50 Berlin companies employing Senior Rust Engineers”)
When it finishes, you stay in the same chat and tell it: “now reformat this in our client report template.” Done.
- Stop being afraid of Claude Code. It is not just for developers.Most non-technical recruiters avoid Claude Code because the name has “code” in it. They assume it’s for engineers. It isn’t. Claude Code is the most powerful operational automation tool on the market right now, and the people using it best are operations leaders, founders, and ops-heavy teams — agency owners and in-house TA leaders who run on weekly reporting, ATS work, and pipeline ops.A real workflow we run every week. Every Monday my team needs progress reports for each active client — funnel data, conversion by stage, candidate movement, close forecast. That used to take half a Monday for one recruiter, every Monday. Now Claude Code does it overnight: logs into our ATS in Chrome, opens each active vacancy and reads funnel data, pulls weekly metrics from our analytics tool, generates a fully formatted report per vacancy in our branded `.docx` template, saves them to a dated Google Drive folder, and posts to Slack: “reports are ready.” The team wakes up to finished reports, adds qualitative notes, and sends to clients. Half the day, gone.
I never wrote a line of code. I described the workflow in plain English; Claude Code asked clarifying questions, I answered, it built and ran the automation. When I want it to change, I tell it in English. It changes.Other things Claude Code now handles for our team in the background:
-
- Pulling new LinkedIn message responses every morning into a tracking sheet
- Updating candidate stages in the ATS based on email replies once a week
- Saving resumes from email, renaming by our convention, sorting into client folders
- Friday: “top 10 candidates at final stage with close forecast” posted to Slack
- Pre-call candidate briefs for the five finalists before a client interview
Claude Code automates the things you currently do by hand. Software is just one of the things it can build. Clicking through your tools, reading your spreadsheets, writing reports, organizing your Drive — equally available.
- Pair Claude Code with a messaging bot. Now your AI team lives in your pocket.Claude Code has one limitation when you first set it up: it runs on your laptop. Which means you have to be near your laptop to give it tasks.There’s a fix. You can configure Claude Code to be controlled through a messaging bot — Telegram in our case, Slack works too. After that you stop being tied to your laptop entirely. Anthropic’s own Claude Dispatch is worth a look here too — I’ve tested it, but our day-to-day setup runs on the Telegram bot, so that’s the route I can speak to from experience.I’m at the gym. I want a Vacancy Overview — status of all active roles with close forecasts — by lunch. I open the bot and type two words: `Vacancy Overview`. The bot triggers Claude Code on my laptop at home; it walks through our nine active vacancies, reads where each candidate sits, who’s at final, what the forecast looks like, and writes a clean summary back to my Telegram. By the time I’m done with the next set, the answer is in the thread.I can dictate by voice. “Prep me for tomorrow’s 11am with [Client X].” The bot launches a Call Preparation skill on my laptop, which pulls context from our `#leads` Slack channel, researches the company online, and sends the brief back to my Telegram.
The mental shift: stop thinking about Claude Code as software running on a laptop and start thinking about it as an AI assistant you message like a colleague. The laptop becomes infrastructure. You stop sitting in front of it.
- Skills are how you turn one-off prompts into a system.The fifth thing — and the one that compounds the value of all four previous — is Skills. A Skill is a saved instruction that triggers a complete workflow. You don’t describe the task again. You just say its name.Three Skills my team uses every week, beyond the obvious Weekly Report and Call Preparation:Database Prep. A favorite of our sourcers, and the one that quietly saves the most hours. A sourcer pulls a contact list from LinkedIn or a scrape — and these always come in messy: email duplicates, LinkedIn URL duplicates, role-based addresses (info@, hr@), broken email formats, mixed scripts in name fields, inconsistent phone formats. Cleaning that used to be an hour of manual work in Google Sheets before every outreach campaign. Now you hit `Database Prep` and Claude removes duplicates, strips role-based addresses, validates email format, normalizes names and phone numbers, splits full name into First/Last, saves the result in the right schema for Lemlist or Reply.io or Smartlead, and returns a report: *”from 1,240 contacts, 987 clean remain. Removed: 134 duplicates, 89 role-based, 30 invalid email.”* Database ready in 30 seconds instead of an hour.
Send to Hiring Manager. For example, a recruiter wraps an interview and instead of an hour of writing, hits `Send to Hiring Manager` and uploads the transcript. Claude reads it end to end, writes a candidate summary in the style and to the standards of the specific client (format, structure, tone, mandatory fields — all already loaded into the Project), saves it as a file in the right Google Drive folder, and drafts the email to the hiring manager with the candidate’s resume and the summary file attached, the right recipient and CCs pre-filled, and the email template for that specific client.
What the recruiter does: opens the draft, carefully reviews the summary (this is the critical step — a write-up about a real person has to be accurate), edits if needed, checks the email body, hits Send. Thirty minutes of work, two minutes of review. The human stays in the loop where the human matters most: the final quality of what the client sees.
- Build your own plugin-agents. The level that separates you from the market.If a Skill is one saved instruction, a Plugin is something more powerful: an agent with several skills, which it can run in sequence to close an entire workflow. Not one action — a complete chain of steps under one umbrella.Two things are worth knowing about plugins.First, there’s an official plugin library from Anthropic — Finance, Legal, Marketing, Sales, Operations, Productivity — out of the box, in a few clicks.Second — and this is where it gets interesting — you can build your own. I personally use custom-built plugin-agents to run my social channels: one agent runs Threads (research → draft → voice-edit → publish → analytics), another runs LinkedIn, another runs my Telegram channel. Each one is a *bundle of skills the agent runs in sequence*, taking a task from idea to published post without me touching every step.
Why this matters for recruiting and sourcing teams: the most repetitive work in our industry isn’t a single action — it’s a *chain* of actions. Custom plugins let you encode those chains:
-
- Outreach Agent — takes a contact list, writes a personalized sequence per persona, splits into waves, pushes to your outreach tool with the right tags and merge fields
- Candidate Pipeline Agent — the full “found → ICP-checked → first message → reply tracked → disqualified or handed to recruiter” loop
- Hiring Manager Update Agent — weekly: pulls pipeline from the ATS, analyzes it, writes the update in the client’s preferred style, adds recommendations, prepares a draft email
- Candidate Onboarding Agent — post-offer: welcome message, document collection, start-date reminders, ping to the client’s HR contact
For the first time, recruiters can build agents that match the actual shape of their work — not generic AI features marketed at us by ATS vendors, but workflows you design around how *your* desk actually runs.
How to actually start — in one weekend
Five steps, in order. Doable over a Saturday and a Sunday.
- Download the desktop app. Cowork — the feature that lets Claude work on your computer — only exists there. The browser version is half the product.
- Build one Project. Pick where you most want help — sourcing, screening, client reporting. Drop in documents that describe how you work. Create 2–3 role-based chats inside.
- Use Deep Research instead of search. Any time you’d have spent two hours Googling, hand the question to Deep Research.
- Get past the fear of Claude Code. Pick one weekly process you do by hand. Describe it in plain English. Let it build the automation. Test once. Then forget it exists — it just runs.
- Set up a messaging-bot remote and Skills. Once one or two automations work, configure a bot so you can trigger them from anywhere. Save the most common ones as Skills.
