| Last Week in HR AI: Issue #8 | Week of August 17, 2026 |
Two UK giants built the same AI tool. The lesson isn't the model.
Plus: Fortitude Re halved time-to-fill, and Stanford's youth employment gap widens to 19%.
|
1 digitaleconomy.stanford.edu2 cxm.world3 pivotnews.ai4 CIO
Ikona's Take on the past week (August 17, 2026) in HR + AI
|
IO
|
Ian O'Keefe
Co-founder & CEO, Ikona Analytics
|
The most useful thing I read this week was not about a model. It was a UK bank with 65,000 employees and a healthcare group with 33,000 independently building nearly the same AI search tool, arriving at the same conclusion, and deleting roughly 60% of their content to get there. Neither team's advantage came from model selection. It came from deciding what was true, retiring what was not, and governing the knowledge that remained. That is a knowledge operations problem wearing an AI costume.
Here is the common thread running through everything in this issue: AI outcomes in 2026 are being decided by the plumbing and the people underneath the models, specifically ungoverned data, undocumented tacit knowledge, and entry-level roles nobody has redesigned. Cloudera found nearly all of 1,500 respondents delayed or killed AI projects over governance and access. Rimini Street is telling CIOs to treat retiring experts as institutional risk on par with technical debt. Stanford's updated payroll analysis shows the employment shortfall for 22 to 25 year olds in AI-exposed roles has widened to about 19%, driven by reduced hiring, and concentrated precisely where AI automates codified knowledge rather than complementing experience-based judgment. Read those three together and the picture is uncomfortable: we are automating the codified layer while the tacit layer walks out the door with the boomers and while we quietly stop hiring the people who used to absorb it by osmosis, earning their stripes, and paying their dues.
So the question I would put on your HR leadership agenda is not which copilot to buy. It is whether you can locate, in structured form, the judgment your HR organization runs on. When we run a diagnostic, 50 structured research interviews typically produces around 2,000 pages of structured qualitative data, and that knowledge was never written down anywhere else. Fortitude Re's self-reported numbers (time-to-fill halved, new-hire attrition down 43%, $2 million in agency fees saved) are real wins worth studying, and note where their CPO drew the line: automate sourcing and scheduling, keep humans on the judgment calls. That line is only defensible if you know what the judgment actually is. If you are wrestling with where yours lives, reply and tell me. I read every response.
| The six to read | The stories that matter for the Office of HR — with our take on each. |
Reviewing submissions to the UK Employee Experience Awards 2026, CXM found that a 65,000-employee bank and a 33,000-employee healthcare and insurance group had separately built almost identical internal AI tools. Both settled on one governed knowledge base, a single search bar in place of chatbots, human escalation for sensitive questions, and each deleted roughly 60% of existing content. Both teams reported that the model choice mattered far less than the governance of the knowledge underneath it. The bank additionally used search telemetry as a live employee listening signal and submitted the tool as a separate responsible-AI entry.
Two 2026 surveys, one from Cloudera and Wakefield Research, one from Google and MIT, report that enterprise data infrastructure designed for human consumption is failing under agentic AI workloads. Nearly all 1,500 Cloudera respondents had delayed or cancelled AI projects because of governance and access problems, and more than half of the Google/MIT sample had paused agent rollouts for the same reasons. Organizations sharing more than 70% of their data with agents report substantially higher trust in agent decisions than those sharing 30% or less.
Also covered by: thejournal.com
In a Q&A with TechTarget, a Rimini Street CIO argues that baby boomer retirements should sit alongside technical debt and cybersecurity on the enterprise risk register. His recommendations include mapping which mission-critical systems depend on a small number of aging experts, running storytelling sessions to capture the judgment that documentation never records, and using AI to mine years of tickets and project records into a searchable repository. The framing treats departing expertise as a quantifiable operational exposure rather than an HR courtesy.
Stanford's Digital Economy Lab updated its 'Canaries in the Coal Mine' analysis using ADP payroll data through mid-2026. The researchers find no broad AI-driven displacement, but the employment shortfall for 22 to 25 year olds in highly AI-exposed roles has grown to roughly 19% below expected levels, up from 15% a year earlier. The gap comes from reduced hiring rather than layoffs, and concentrates in work where AI automates codified knowledge instead of complementing tacit, experience-based expertise.
Reinsurer Fortitude Re rebuilt its talent acquisition function around tools including SeekOut, LinkedIn Recruiter, and Greenhouse, reporting a 50% reduction in time-to-fill, a 43% drop in new-hire attrition, and $2 million saved in agency fees. CPO Denise Nichols told HR Executive that automation should own sourcing and scheduling while humans keep decisions on culture fit and collaboration. She also contends AI is creating demand for new roles rather than only reducing headcount. The figures are self-reported and have not been independently verified.
BBC reporting contrasts tech executives' predictions of shorter working weeks with accounts from employees at OpenAI, Anthropic, Meta, and Google describing 70- to 90-hour weeks, abrupt team reassignments, and punishing release sprints. A UC Berkeley study and commentary from an MIT scholar suggest AI time savings are absorbed by new tasks and by the oversight work agents create rather than converted into reduced hours.
Also covered by: webpronews.com
Most HR functions hold plenty of system data and almost none of the tacit knowledge that explains how work actually gets done, and that gap is why generative AI projects stall inside HR. [The piece argues](https://www.ikonaanalytics.com/insights/the-hidden-data-deficit-that-will-sink-your-hr-ai-strategy) that owning the context layer is the next strategic claim available to analytics leaders.
Both UK teams landed on the same finding from opposite industries: the model matters less than the knowledge governance underneath it. That should be encouraging, because governance is something you can actually control this quarter, unlike the model roadmap of whichever vendor you signed with. The harder truth in the BBC reporting is that time saved has a way of getting absorbed rather than returned, so if you want AI to buy your team capacity, you have to decide in advance what that capacity is for. Leading HR through this moment is genuinely hard, and the leaders who keep making deliberate, documented moves (even small ones) are the ones who will look back in a year glad they did. See you next week.
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.