| Last Week in HR AI: Issue #17 | Week of October 19, 2026 |
Pitching AI as an efficiency tool made employees engage less
Paralegals given time to explore, training, and peer sharing used the tool 58% more.
|
1 fromdayone.com2 Human Resources Director3 hai.stanford.edu4 clo100.com
Ikona's Take on the past week (October 19, 2026) in HR + AI
|
IO
|
Ian O'Keefe
Co-founder & CEO, Ikona Analytics
|
Most HR functions I talk to are about to put AI agents on top of judgment they have never written down. That is the through-line of this week's reads: AI performs only as well as the business context and redesigned workflow underneath it, and building that context is HR's work before it is a technology project.
The evidence lined up uncomfortably well. In Optro's survey of 417 governance, risk, compliance, and audit leaders, 34% of organizations had acted on an inaccurate agent output. The causes were weak data and agents dropped into workflows nobody redesigned. Microsoft's AI read an internal network system's name as being about birds, which is why its engineers insist that people supply the meaning. Zapier's rating assistant works for the opposite reason: written standards and past calibration outcomes exist to check managers against.
The harder point is oversight. A reviewer with no time to evaluate an agent's action produces a signature that looks like a control. Angie Ng at Zapier has the right rule: oversight should scale with the cost of an error. That is an architecture decision you make up front. So before your next agent pilot, ask where your rating bar, your job architecture, and the real logic of your core processes actually live. If the answer is three people's heads, that is your first build. I'd like to hear where you landed. Reply and tell me, or reach out if you want to work through it together.
| The six to read | The stories that matter for the Office of HR — with our take on each. |
Zapier's people team has built an artificial intelligence (AI) assistant that reviews each manager's written rating justification against performance standards and prior calibration outcomes. It flags mismatches and shows where individual managers' rating bars diverge, while people make every rating decision. Director Angie Ng says oversight should scale with the cost of an error, so coaching drafts always go to a human resources (HR) partner first. The article ties the work to European Union (EU) AI Act employment rules, which it dates to December 2027.
Speaking at a From Day One conference, TIAA's chief people officer said the firm treats AI disruption as a reason to move current employees into new roles instead of hiring externally. Recruiters were moved to internal placement, the company launched an employee program called Elevate and an in-house AI tool, and department leaders were given quarterly mobility targets. She reports an internal mobility rate of 45%, compared with an industry average of 27%.
TP's global learning and development (L&D) director describes a multi-year program to make skills the organizing logic of a 500,000-person business. The work starts from a base taxonomy built by Mercer, moves through job architecture and labor market data, and ends with technology connecting skills to hiring, development, performance, and mobility. She measures L&D on attrition, speed to hire, and speed to competence. AI is used for vendor analysis and personalized development.
|
From Ikona
Your AI is only as good as the system around it
We diagnose how work and knowledge actually flow through your HR function and capture the judgment that lives in people's heads. Then we structure it into standards and definitions an AI system can stand on. Talk to us about yours → |
Microsoft's internal information technology (IT) group lays out how it is preparing enterprise data for agentic AI, treating readiness as a standing capability rather than a single cleanup. Data is scored on quality, governance, context, lineage, interoperability, and observability, and leaders plan to track readiness monthly. Engineers say people still have to supply business meaning, pointing to an AI that read the name of an internal network system as being about birds. AI drafts data descriptions and flags stale data, and humans verify the results.
Optro surveyed 417 governance, risk, compliance, and audit leaders and found that 34% of organizations had acted on an inaccurate agent decision or output. Among organizations deploying agents, 25% reported a control failure and 24% reported customer-facing impact. The causes cited include weak data quality and agents added to workflows that were never redesigned. Human reviewers can override agents, but they often lack the time to evaluate actions properly.
Also covered by: Human Resources Director, news.sap.com
Researchers from Stanford, the University of Virginia, and Dartmouth followed generative AI rollouts at a law firm, an advertising agency, and an IT services firm. Employees who heard AI pitched as an efficiency play engaged less than those told it would enrich their work. At the law firm, paralegals who received exploration time, training, and peer sharing used the tool 58% more. The authors recommend swapping volume metrics and token-consumption tracking for measures of value.
Also covered by: jbiznews.com, SHRM
Many enterprise HR functions can report their metrics but cannot trace why things break. The piece names the missing capability organizational literacy, the ability to understand the organization as a connected system, and outlines what it takes to build that diagnostic muscle.
Time showed up twice in this week's evidence, and both times it pointed the same way. In Optro's survey, human reviewers could override an agent but often lacked the time to evaluate what it had done; at the law firm Stanford studied, paralegals who got exploration time, training, and peer sharing used their tool 58% more. Whose hours get protected is a call HR makes, and that turns Angie Ng's rule into something you can write into a project plan: wherever a wrong output could cost someone a rating or a role, budget the reviewer's hours right next to the license fee.
If your team is wrestling with workforce intelligence, AI readiness, or a transformation that has to land, we'd love to help. Reach out and tell us what you're working on.
Get in touch →Last Week in HR AI
Practitioner perspectives and diagnostic frameworks, delivered the week they publish. Be the first to know when new research drops.
No spam, ever. Unsubscribe anytime.