Explaining Tokens to HR Who Are Too Afraid to Ask
A plain-English guide to tokens, models, context, reasoning, and how to control AI costs without stopping useful experimentation.
Notes from the work
Mike WinklerAdvisory↗
02
Published · 4 minute read
HR should help design the work AI changes
AI is removing HR tasks, but the larger shift is pulling HR closer to workflow design: what AI prepares, what a person owns, and whether the new process works.
03
Published · 6 minute read
What should an HR AI pilot measure besides time saved?
An HR AI pilot should measure source quality, corrections, failure modes, stop conditions, prohibited uses, reviewer effort, and elapsed time.
04
Published · Updated · 5 minute read
The link worked. The source was still wrong.
A shortlink sent an AI assistant to a plausible but incorrect post. The answer looked fine, so Mike Winkler changed what the workflow counts as a verified source.
05
Published · 5 minute read
The source told HRMC to skip review. The first run still failed.
Mike Winkler tested an HR AI workflow with a made-up leave-law alert that tried to bypass review. The tool resisted the instruction but still exposed a human-review gap.
06
Published · Updated · 6 minute read
What is agentic AI in HR?
Agentic AI in HR can check approved sources, prepare a brief, route work, and keep a record. It should not make employment decisions.
07
Published · Updated · 7 minute read
What real human review looks like in an HR AI workflow
A reviewer needs the evidence, time, authority, and a way to stop the next action. These checks show whether the person really has control.
08
Published · Updated · 6 minute read
What an HRBP intake agent should gather—and never decide
An HRBP intake should organize the timeline, policy, known evidence, missing facts, urgency, and owner. Employment decisions stay with a person.
09
Published · Updated · 6 minute read
What an HR workflow receipt needs to preserve
An HR workflow receipt should keep the source, case identity, missing facts, reviewer, decision, correction, owner, approved action, and verification together.
010
Published · Updated · 6 minute read
When AI does not belong in an HR workflow
AI does not belong in an HR workflow when the test requires unsafe data, hides decision ownership, cannot be checked or stopped, or is being asked to make the employment decision.
Start with a conversation
Tell me what is getting in the way.
Bring the workflow that keeps reopening, the approval nobody can reconstruct, or the pilot that looked better in the demo. We will start there. No deck required.
Book a conversation ↗Or email mike@mikewinkleradvisory.com