| Last Week in HR AI: Issue #12 | Week of September 14, 2026 |
They skipped the recruiting AI and still cut 27 days
Uniper fed Copilot the process context it already owned instead of buying a tool. Screening fell 13 days too.
|
1 forsmile.jp2 diginomica.com3 medcitynews.com4 Josh Bersin
Ikona's Take on the past week (September 14, 2026) in HR + AI
|
IO
|
Ian O'Keefe
Co-founder & CEO, Ikona Analytics
|
Three of this week's stories are the same story told in different currencies. Panasonic Connect started building its corpus in 2023 before it let agents near it, and reported 788,000 hours back in FY2025, about 3.4% of total work time. Uniper skipped buying a recruiting AI and fed Copilot the process context it already had in Celonis; hiring moved 27 days faster. Hong Leong Investment Bank wired specialized agents around an orchestrator, cut credit work from days to minutes, and caught a spreadsheet error nobody had spotted in years. The through-line is the context layer: your data platform, your documented process, your captured expertise, is what decides whether agents give back hours, days, or nothing.
Cleveland Clinic is the one I would put in front of a CFO. Over 4,000 ambulatory clinicians, 60% saying the ambient scribe makes them more likely to stay in practice, adoption voluntary rather than mandated. The return landed as retention, not headcount reduction. That is a harder number to book and a much easier one to defend.
Then Maersk and the new BLS projections complicate all of it. If AI is removing the junior tasks where expertise used to form, and the labor force grows 3.5% this decade while output grows 22%, then your foundations work solves the agent problem and leaves the supply problem sitting untouched. Capturing what your experts know stops being an efficiency play and becomes a hedge against a bench you are no longer building.
Tell me where you think I have this wrong. I read every reply.
| The six to read | The stories that matter for the Office of HR — with our take on each. |
Drawing on Panasonic Connect's own disclosures, a tech blog reconstructs three years of the company's ConnectAI rollout across roughly 11,800 Japanese employees. FY2025 savings reached 788,000 hours, about 3.4% of total work time, up from 448,000 hours the year before. The sequence matters: a data platform called the Connect Corpus was built starting in 2023, RAG access followed in 2024, and a Snowflake-based drawing-comparison agent arrived in 2026. The piece's argument is that data readiness came first and agent deployment second.
German energy company Uniper connected Celonis process-mining context to Microsoft Copilot to run HR AI agents in recruiting. Instead of purchasing a dedicated recruiting AI product, its Center of Excellence built on the Microsoft and Celonis licenses already in place. Feeding Copilot a structured business context model moved the agents from generic responses to specific ones. Uniper reports time-to-sign down 27 days and screening time down 13 days, with the agents deliberately kept out of candidate selection decisions.
Also covered by: itbrief.com.au
Hong Leong Investment Bank has assembled what it calls an institutional brain: a set of specialized AI agents coordinated by an orchestrator named Mala, first deployed in risk management and since extended to sister firm Hong Leong Asset Management. Credit assessments and stress tests that took days now run in minutes, and the agents surfaced a spreadsheet error that had gone unnoticed for years. HLIB leadership describes the goal as retaining expertise that would otherwise walk out with departing staff. Cost reduction is framed as a secondary benefit.
|
From Ikona
Your agents are only as good as the context you feed them
We diagnose how work and knowledge actually move through your HR function, then capture the judgment sitting in your people's heads. That becomes a structured knowledge store your AI can stand on. Talk to us about yours → |
Cleveland Clinic research published in npj Health Systems tracked the deployment of Ambience Healthcare's ambient AI scribe to more than 4,000 ambulatory clinicians. Sixty percent of users said the tool made them more likely to stay in practice, with 97% reporting satisfaction and 70% still using it after a year. Clinic leaders credit voluntary adoption and fast, flexible training rather than a mandate. The study frames the outcome primarily in retention terms.
Maersk's global head of talent management, Anish Lalchandani, contends that most reskilling programs fail because they are built around what the organisation needs rather than what employees want to become. He argues fixed skills taxonomies are outdated by the time they ship and proposes a live skills canvas in their place. He also warns that AI is removing the junior tasks through which expertise has traditionally been built, and says global capability centres, not relocation, are now where cross-market leaders are formed.
Josh Bersin reads the new BLS Employment Projections for 2025 to 2035 and finds the US labor force growing just 3.5% over the decade while GDP rises 22%. That gap implies nearly 2% annual productivity growth and points to labor shortages rather than AI-driven surplus. Healthcare and home care account for most job growth, while administrative and low-touch roles contract. Bersin calls the pattern a Superworker shift in which AI upgrades most jobs instead of eliminating them.
The argument is that model choice is a depreciating advantage: today's frontier model is next year's commodity, while the structured knowledge layer underneath it is the asset that actually compounds.
Maersk's Anish Lalchandani says the advantage is having the fastest map rather than the most complete taxonomy, and after reading this week I think he is half right. The map only moves fast because someone wrote down the territory first, which is the three years Panasonic spent on its corpus before an agent touched anything. What none of these six organizations solved is the other half: where the next generation of experts comes from once the junior work is gone. That is the question I am carrying into 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.