I Went Down an AI Rabbit Hole and Built Something. Here’s What Came Out.

A few weeks ago I got stuck on a question all talent leaders are grappling with: where should I actually start with actual AI automation in my TA operation?

Not “what’s theoretically possible” — we’ve riffed on that plenty. I mean the tactical version of the question. The one that matters on a Monday morning. If I’m going to invest time building automation workflows, which problems are actually worth solving first?

I couldn’t find a satisfying answer anywhere. So I started mapping it out myself. That turned into a late-night build session. Which turned into a (primitive) prototype I’m now releasing publicly, because I’m pretty sure I’m not the only person wrestling with this.

---

What I Was Actually Trying to Solve

Here’s my frustration with most AI-in-TA content right now: it operates at the “here’s what AI could do” level. Which is fine. But it doesn’t help you figure out where to start when you have broken workflows, limited time, and a team that’s already behind the adoption curve.

What I wanted was something diagnostic. A way to score my own workflows — not just by how painful they are, but by how much fixing them could actually change my outcomes.

Because those are two very different things.

Processing offer letters is tedious. But automating it doesn’t make you a better recruiter. It just makes paperwork slightly less of a drain. Compare that to sourcing and pipeline management — where the inefficiency isn’t just costing you time, it’s costing you candidates. That’s a fundamentally different problem with a fundamentally different ROI.

I wanted a tool that made that distinction. So I built one using Claude to suggest tailored automation streams in Claude Cowork (meta, I know - no, not that Meta).

---

What It Is (Honest Version)

I want to be upfront: this is a primitive prototype. Version 0.1. It works, it gives you real output, but don’t show up expecting to re-engineer your entire TA systems based on its output. It’s a directionally sharp, opinionated diagnostic that takes about five minutes and leaves you with an actual action plan and suggested prompts to get started. If you’re just learning about Claude Cowork in this post, be sure to pop over to this article to learn more from Anthropic.

The TA Automation Audit walks you through seven workflows:

• Req Intake & Job Descriptions

• Sourcing & Pipeline Management

• Candidate Communications

• Interview Scheduling & Coordination

• Offer Management & Documentation

• Reporting & Analytics

• Onboarding Handoff

For each one, you rate three dimensions:

• Time Drain — how much weekly time is your team actually spending here?

• Repetition Rate — is this work predictable and templated, or does it vary enough that automation gets messy?

• Automation Upside — this is the one that matters most to me. Not “how painful is this?” but “how much would fixing this change what we’re able to do?”

The defaults on that third slider are opinionated. Sourcing & Pipeline is pre-set to 5. Offer Management is pre-set to 2. That’s intentional. You can change them — your context is yours — but the starting point forces the right conversation.

Your three scores multiply together (max 125). Anything above 50 is a Start Here. 20-49 is Phase 2. Under 20 goes in the Monitor bucket — not because it doesn’t matter, but because the automation math doesn’t work yet.

---

What You Actually Get

Here’s what I’m most proud of, even in prototype form: the output is specific.

Not “consider automating your candidate communications.” But here’s the Cowork workflow, here’s the folder structure, here’s the actual prompt to drop in today. Like this one from the Sourcing & Pipeline section:

“Using the sourcing list in [file], clean and deduplicate this list against our existing candidate database in [ATS export]. Flag any candidates we’ve previously contacted, note their last touchpoint date, and rank the remaining names by fit score based on our ideal candidate profile in [ICP doc]. Output a clean, prioritized outreach list with a recommended personalization hook for each candidate based on their background.”

That’s not a concept. That’s a workflow you can run this week.

Every recommendation comes with use cases for that specific workflow and step-by-step instructions for building it in Cowork — watched folders, trigger logic, sample prompts. The whole shabang.

---

Why I’m Putting This Out Now

Mostly because I think practitioners need to share more concrete tools and techniques and fewer hot takes when it comes to practical, tangible applications of AI.

Every week, I talk to talent leaders who know AI automation is coming, want to get ahead of it, but don’t know where to start. They’re not asking philosophical questions. They’re asking: what do I do Monday morning?

This is my attempt at an answer. For myself, initially. And now with a framework for finding your own.

I also know that releasing it publicly is the fastest way to improve it. There are things I already want to build in — team size weighting, ATS-specific recommendations, more granular workflow breakdowns. But I need to know what’s actually useful to people before I go further down that road.

---

Give It Five Minutes

The tool is free and lives on larsschmidt.com (it’s HTML that needed to be housed somewhere). No login, no form to fill out, no email required or gated lead gen. Just an open-source MVP, I invite you to help me make it better.

https://larsschmidt.com/ta-ai-automation-audit

Run through all seven sections honestly. The sliders are only useful if you’re real with them.

Then — and this is the part I’m genuinely asking for — tell me what’s wrong with it. What’s missing? What felt off? What would make the output more useful for how your team actually operates?

Reply here, find me on LinkedIn, or just shoot me a note. I’m not being polite when I say I’ll read everything. I actually will.

This is version 0.1. Where it goes from here depends entirely on what you tell me.

Thanks for reading! ✌🏼