👋 Hey,

Welcome back to PromptMates Weekly.

This week: Glen Cathey made the case for eliminating annual interview training, we share a 1-2 punch to eliminate applicant spam, and Workday dropped $700M on Sana.

🗓️ Tuesday, March 25 at 5:00 PM CET

Luke Eaton, Director of Recruiting at Ava Labs, demos live AI workflows across headcount planning, predictive sourcing, kick-offs with hiring managers, and talent intelligence.

Not slides - a live walkthrough of real systems running in production. Audience Q&A included.

Save your spot —>

The most powerful agent for inbound recruiting.

Metaview just launched Application Review. The AI agent that evaluates every applicant against your team's hiring standards. Instantly surfacing the strongest matches.

No auto-rejections. Your definition of great, applied at scale. And it lives inside the same platform where you already source, interview & hire.

→ Humans define their Ideal Candidate Profile
→ The agent flags top-fit candidates instantly
→ Humans decide who moves forward
→ The agent keeps improving from your feedback

Check it out here —>

Category: Application Defense Strategy

Robbie Simpson breaks down a simple strategy to stop application spam, where candidates use tools like LazyApply and AI Hawk to blast thousands of auto-applications daily. The industry solution is expensive AI-filtering tech that ruins the experience for real candidates. There's a simpler approach that breaks the bots without breaking your candidate experience.

When it works: High-volume roles getting slammed with spam applications where you need immediate relief. Works best when you have control over your careers site code and can configure how your ATS displays job parameters. Most effective for companies using standard ATS platforms like Greenhouse (gh_jid), Lever (posting_id), or Workday (requisitionId) that bots are trained to recognize.

When it doesn't: Low-volume hiring where application spam isn't a major problem. Won't work if your recruiting team relies heavily on LinkedIn Easy Apply for candidate volume. Likely only effective for a few months before bots adapt. Requires technical access to modify your careers page code if you can't hide or obfuscate ATS identifiers like ?gh_jid= parameters in Greenhouse, the second step won't work.

Bottom line: Two configuration changes-turning off LinkedIn Easy Apply and hiding ATS identifiers in your careers page code-force manual action that real candidates will do but bots are programmed to avoid. Not a permanent solution, but it makes your company computationally expensive enough that bots move on to easier targets.

Try it →

Your hiring managers forget 90% of interview training within a week. That is not a training quality problem - it is a fundamental mismatch between event-based learning and how humans actually develop skills. Glen Cathey breaks down why continuous AI coaching before, during, and after every interview is the only approach that works at scale.

Pull your interview data from the last 6 months. How many hiring managers completed training? How many interviews happened more than 30 days after that training? For most orgs, the answer exposes the core problem: managers interview infrequently, and by the time they do, they have forgotten 79% of what they learned. That is your baseline - and it explains why structured interviewing adoption is inconsistent no matter how good your training content is.

Create an AI prompt that takes the role description, candidate background, and competencies being assessed as inputs. The output should generate tailored behavioral questions, remind the interviewer what to listen for in responses, flag potential bias triggers for this specific candidate-role combination, and walk through evaluation criteria. This is the pre-game preparation - every athlete gets it before they compete, your hiring managers should get it before every interview.

Configure an AI assistant that hiring managers can reference during the interview. It should suggest follow-up probes when a candidate gives vague answers, prompt the interviewer to ask for specific examples, alert when too much time is spent on one area, and flag when questions veer into legally risky territory. Think of this as the coach on the sideline making real-time adjustments. Critical distinction: this coaches the interviewer, not evaluates the candidate.

After each interview, run the interviewer's notes and any transcripts through an AI analysis. Did questions stay on-topic? Were conclusions supported by behavioral evidence or gut feeling? Were there patterns suggesting bias? Compare this interview to the manager's previous interviews - are they improving or regressing? This is the film review session that drives continuous improvement.

Log every pre-brief, real-time assist, and debrief into a dashboard. After a month, you will see which managers consistently follow structured interviewing and which fall back to pet questions and gut instinct. Use this data to target additional support where it is needed - not another training session, but more intensive coaching for specific skill gaps.

💡 Pro Tip: Start by coaching interviewers, not evaluating candidates. That distinction matters enormously for legal and ethical defensibility. You are not using AI to make hiring decisions - you are using AI to help human decision-makers get better at their job. Frame it that way internally and adoption resistance drops significantly.

⚖️ TestGorilla: AI Fluency Is Now About Accountability, Not Speed

Early AI adopters rewarded prompt engineering and output velocity. Then datasets got pushed into models they shouldn't have touched, compliance checks got skipped, and decisions made at AI speed lacked human judgment. The correction is underway. Companies are now hiring for five pillars: Digital Agility, Systems Thinking, Ethics, Human-AI Collaboration, and AI Literacy. The weight given to each varies by sector and data sensitivity. A marketing team and a genomics lab aren't hiring for the same AI fluency.

See the framework →

🔧 The AI Laundry: Why 'Talent Engineer' Roles Are Scoped Wrong

xAI, Metaview, and a growing list of startups are hiring Talent Engineers to build automations and integrate APIs. They found the problems weren't technical. Candidates waiting eight weeks for feedback wasn't a system problem; hiring managers just didn't have hiring in their KPIs. Final-round panels of eight people were a leftover from when the company was 40 employees. The recruiter's all-in-one HR system couldn't show her what needed attention, so she built a fake pipeline stage called 'to do.' Most companies need ownership fixes, not API workarounds.

Read the breakdown →

🔬 n8n: Three Patterns for Human-in-the-Loop AI Workflows

Capability without oversight is a liability. This production AI playbook shows you how to build human oversight into AI workflows without killing the speed that made automation worthwhile. Three patterns: inline chat approval, tool call approval gates, and multi-channel review workflows. Each pattern includes when to use it and when not to. The VP-of-Sales-90%-discount-email scenario is painfully real.

Get the patterns →

💡 Microsoft Copilot Cowork: Task Execution Layer Built Into M365

Most AI tools are great at the first 20 percent, then they hand it back to you and the real work begins. Copilot Cowork collapses that loop. It's not a chat upgrade; it's a task-execution layer that runs across your actual M365 apps and files. Calendar triage, meeting prep, company research, launch planning. Built in partnership with Anthropic, with Claude powering parts of the execution layer. Currently in Research Preview, broader access rolls out through Microsoft's Frontier program in late March.

See what it does →

Microsoft's control plane for tracking what agents exist in your org, who's using them, what data they touch, and where the risks are. More than 80% of Fortune 500 companies already have agents running, but most lack mature governance.

You can't govern what you can't see. If your recruiting team is running agents for candidate outreach, interview scheduling, or sourcing research, IT needs visibility into what those agents can access. Agent 365 treats agents like digital employees - with identities, permissions, and audit trails. That's the infrastructure layer that makes agent adoption defensible to your CISO.

See the control plane →

Workday now has an agent interface that lets you query, report, and run transactions directly against your Workday data. Built on Sana Labs' agent platform (the tech behind Josh Bersin's Galileo), it brings voice generation, meeting analysis, and workflow automation into one experience.

If you're a Workday shop, this changes how your team accesses recruiting data. Instead of clicking through five screens to pull time-to-fill by department, you ask. Instead of exporting CSVs to analyze offer acceptance rates, the agent runs the report and explains the trend. Workday's security rules are baked in, so each user sees only what they're allowed to see - but the interface is conversational, not transactional.

Read the strategy →

Charlie Guo from OpenAI laid out three phases of AI coding: autocomplete, pair programming, and agent delegation.

Phase 3 is where you manage multiple agents across large workflows. You're the manager. They're the team. This isn't a recruiting article. But the framework applies directly.

Most TA teams are stuck in Phase 1 (autocomplete for job descriptions) or early Phase 2 (ChatGPT as a writing assistant). Phase 3 means delegating entire workflows: an agent that monitors your ATS, identifies drop-off patterns, drafts outreach sequences, A/B tests subject lines, and reports back with recommendations.

You review. You approve. You don't execute.

The capability jump that made this possible for engineering-GPT-5.2 to GPT-5.4 is the same jump available to recruiting tools today.

The question isn't whether agent delegation works. OpenAI proved it does at scale. The question is whether your team is structured to manage agents instead of doing the work yourself. Most aren't.

If you handed an agent your entire sourcing workflow today, what would break first-the agent's capabilities or your team's ability to manage it?

If you're building AI tools for your recruiting team, these patterns are your blueprint. Start with the human oversight layer first, then automate around it.

See the full playbook →

Go deeper

If you want more builds, prompts, and workflows like the ones in this newsletter, that's what the PromptMates community is for — templates, live sessions, and a space to ask questions when you get stuck.

Here's to working smarter, not harder, and actually enjoying the tech that's supposed to help us.

Join the community →