| Last Week in HR AI: Issue #16 | Week of October 12, 2026 |
Your AI upskilling plan needs builders now, not prompt training
Workday's data: demand for basic prompting fell after a January 2026 peak, while workflow automation rose.
|
1 brandsit.pl2 news.beaconturn.com3 govtech.com4 irishtimes.com
Ikona's Take on the past week (October 12, 2026) in HR + AI
|
IO
|
Ian O'Keefe
Co-founder & CEO, Ikona Analytics
|
Two of the largest HR platforms just told you where they think the value in agentic AI sits: in the context layer underneath the agent. SAP agreed to buy TechWolf, the skills-inference company Workday had already deployed across its own 20,400 employees, while Workday builds on its own transaction data. My read of the week: the vendors are fighting over who owns the context your agents run on, while most HR functions have not done the work to make that context worth owning.
The evidence on that second half is uncomfortable. Workday's own report shows internal moves declining at most employers, which is exactly the outcome a skills layer is supposed to fix. Mercer found dissatisfaction with HR tech stacks rose from 28% to 39% in a year. The operators getting results did the unglamorous part first. South Carolina made data quality and tagging prerequisites and funded a chief data officer to own them, and HelloFresh's Syed Ali Abbas tells peers to fix data before anything else.
The résumé study shows what skipping that step costs. A Copilot agent screening 178 applications favored polished, implausibly strong résumés that looked AI-written. In a separate simulation, candidates using the recruiter's own model were 23% to 60% more likely to be shortlisted.
That is agent judgment resting on inputs nobody examined. Before your next human capital management (HCM) contract, ask three things: where does your agent's skills data originate, who has validated it, and what happens if that partner is acquired? Reply with your answer, or reach out and we'll dig in together.
| The six to read | The stories that matter for the Office of HR — with our take on each. |
SAP has agreed to acquire TechWolf, a skills-inference company that Workday had already rolled out to its own 20,400 employees. Lance Haun reads the deal as a sign that the large HR vendors are splitting into two camps over where agent context should come from. In his reading, Workday builds on its own transaction data and buys points of access, while SAP purchases context layers outright, with ServiceNow, Microsoft, Gloat, and Celonis competing in the same space. His advice to human capital management (HCM) buyers is to ask where an agent's skills data comes from and what happens to it if a partner gets acquired.
Workday's October 2026 report, titled The Adaptability Advantage, draws on customer workforce data, requisitions from roughly 550 employers, and two surveys of workers. Leaders lean toward using artificial intelligence (AI) to raise output from current staff rather than to reduce headcount. Demand for basic prompting skills dropped after peaking in January 2026, while demand for building AI tools and automating workflows climbed. Internal moves fell at most employers, applicants per hire went up while time to fill held steady, and employees describe a gap between knowing which skills they need and getting support to build them.
Also covered by: workday.com
Worth translating this one from Polish as it's a great collection of sources. A Brandsit analysis contends that HR technology is past its peak of tool buying. Okta data puts the average organization at around 101 business applications, and Mercer found dissatisfaction with HR tech stacks climbing from 28% to 39% in a single year. Fosway reports that European buyers now put integration ahead of new purchases, and examples from Allegro, Unum, and Veepee show savings from consolidation. Deloitte cautions that few firms actually orchestrate their data, so AI agents often work through existing disorder at greater speed.
|
From Ikona
Your agents are only as good as your context
We interview the people who run your HR function and map how work, data, and judgment actually move through it. You keep that as a structured, queryable knowledge asset your AI programs and vendor decisions can stand on. Talk to us about yours → |
One year after launching its Bradley AI assistant, South Carolina tracks about 250 AI use cases and treats data quality, tagging, and sharing across agencies as prerequisites for useful results. The state funded a chief data officer role with AI responsibilities and is modernizing legacy applications, growing shared services, and adding cybersecurity controls. Chief information officer (CIO) Nathan Hogue said an internal Copilot and ChatGPT pilot added roughly 15% to his weekly capacity. He also acknowledged that leaders are still questioning whether the effort pays off.
Syed Ali Abbas, vice president of people at HelloFresh, says HR should point AI at reinventing the whole business rather than at administrative fixes such as speeding up onboarding. He describes combining AI with robotics and supply-chain software in production centers and widening recipe variety. A 350-recipe pilot was scaled back after it put strain on systems and quality. His advice to fellow HR leaders is to fix data first, let early adopters lead before pushing wide adoption, and keep HR out of leading the technology itself.
Addleshaw Goddard ran a Microsoft Copilot agent over 178 applications for an HR information technology (IT) role, and the agent's shortlist differed from the one its human reviewers produced. The agent favored polished, implausibly strong résumés that appeared to be AI-written. A separate working paper simulating hiring across 24 jobs found candidates who used the same language model as the recruiter were 23% to 60% more likely to be shortlisted. Instructing models to disregard where a résumé came from and running several models cut that bias by more than half, and humans made the final hire.
Argues that treating AI delay as the cautious choice hides four compounding costs: tenured knowledge leaving, shadow architectures spreading, vendor commitments signed against an unknown current state, and foundation work that grows harder and more expensive over time.
Nathan Hogue says South Carolina's leaders are still asking whether "the juice is worth the squeeze," and SAP answered that question for itself when it agreed to buy TechWolf. Vendors pay for context layers because the context is where agent value sits. The résumé study shows how quickly that layer improves once someone looks at it: telling the models to ignore a résumé's origin and running several models cut the bias by more than half. That is the part of the context fight your own team controls, and it costs far less than an acquisition.
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.