| Last Week in HR AI: Issue #10 | Week of August 31, 2026 |
Verizon's Gemini answers most calls. The data work came first.
Farmers freed 16.4M agent hours. Meta scrapped its second layoff wave. The gap is foundations.
|
1 theglobeandmail.com2 thenextweb.com3 insurancebusinessmag.com4 theaiinnovator.com
Ikona's Take on the past week (August 31, 2026) in HR + AI
|
IO
|
Ian O'Keefe
Co-founder & CEO, Ikona Analytics
|
Two of this week's stories are about AI failing, and neither failure was about the model. Meta modeled team cuts of up to 60% under a plan called Project OT, executed 10% in May, then scrapped the second wave when its own data showed the agents were not delivering. Starbucks pulled an inventory tool out of 11,300 stores because reflective surfaces confused the computer vision, seasonal packaging needed weeks of retraining, and a 1990s IBM backend could not move real time data. Nothing exotic. Plumbing.
Now the other four. Verizon has Gemini answering most inbound calls and chats, sitting on years of consolidating its data into one layer. Farmers merged 300,000 documents and five lookup systems into a single interface, resolved more than 60% of questions that used to reach call centers, and freed 16.4 million agent hours. Adecco cut time to first candidate submission by about 40%, and learned that automation cannot fix a broken process; it just moves the bottleneck downstream. The through-line is simple: the foundation work, not the model, is still the deciding variable.
That should change what you ask for this quarter. If the most advanced artifact in your AI program is a vendor demo, you are Starbucks in month one, not Verizon in year three. The diagnostic question is narrow: which of your HR processes has data clean enough and rules explicit enough for an agent to act on, and who on your team can answer that today? If you know, I would like to hear it. If you don't, reply and let's dig in.
| The six to read | The stories that matter for the Office of HR — with our take on each. |
A Reuters investigation describes an internal Meta effort called Project OT, a 2026 push to make the company "AI native." Executives modeled team reductions of up to 60% and planned two layoff waves. Meta carried out a 10% cut in May but abandoned the second wave after internal data showed AI agents were not producing the productivity gains that had been assumed. Staff pushed back, reading the AI framing as cover for headcount decisions.
Also covered by: Human Resources Director
Starbucks shut down an AI inventory tool built by 30-person startup NomadGo after it had already been rolled out to all 11,300 North American cafés. Fast Company's reporting traced the failures to ordinary causes: reflective surfaces confusing the computer vision, seasonal packaging requiring weeks of retraining, and a 1990s IBM backend unable to move real-time data. Baristas found out six weeks after the vendor did. The CTO who championed the tool resigned shortly after full rollout, and NomadGo cut much of its staff within days of losing the account.
Google Cloud and Verizon announced a broad partnership deploying Gemini Enterprise across customer service, network operations, marketing, security, and productivity tools. Gemini already handles most of Verizon's inbound calls and chats, built on a multi-year consolidation of Verizon's data into a single unified layer. The piece connects the deal to Google Cloud's record Q1 2026 growth and argues the data foundation work, not the AI layer, was the actual prerequisite.
Also covered by: cryptobriefing.com, telcomagazine.com
|
From Ikona
Your AI program is only as good as the plumbing
We diagnose how work and knowledge actually move through your HR function, capture the judgment living in people's heads, and turn it into a structured knowledge store an AI system can stand on. Talk to us about yours → |
Adecco SVP Pierre Matuchet outlines the firm's deployment of AI agents, built on Salesforce's Agentforce, across pre-screening, talent pooling, and redeployment. The agents run continuously and have cut time to submit first candidates by roughly 40%, while final interviews remain human. Matuchet notes two lessons: automation can push bottlenecks downstream, and it cannot repair a process that is already broken.
Farmers Insurance reports that an AI program reduced routine servicing tasks for its 8,000 agents and roughly 20,000 support staff by 35%, freeing about 16.4 million hours annually for sales and client work. A large language model tool, askfarmers.ai, consolidated more than 300,000 documents and five separate lookup systems into one chat interface, resolving over 60% of questions that previously reached call centers. Leadership says team size stayed flat as the portfolio grew and describes the work as augmentation rather than headcount reduction.
Rebecca Port, Okta's chief people officer since January 2026, describes a commitment to give employees back 100,000 hours in 2026 by simplifying HR processes. Completed changes include replacing heavy mid-year reviews with short check-ins (9,000 hours saved), cutting engagement surveys from 40 questions twice a year to five once a year, consolidating manager training, and reducing clicks in HR systems. Port ties the effort to Okta's agentic AI plans and describes a shift from "workforce planning" to "work planning."
Argues that meetings were always context transfer machines, and now that software is making the calls that used to happen in those rooms, HR is handing decisions to systems that carry almost none of what a person would have brought. Getting that context into machine-readable form is the foundation work this week's winners already did, and HR starts further behind than any function in the building.
The detail I keep turning over is not the reflective surfaces or the 1990s backend. It is that Starbucks baristas found out the tool was dead six weeks after the vendor did, which tells you the information channels were broken long before the computer vision was. Farmers spent its effort collapsing 300,000 documents and five lookup systems into one place, and the 16.4 million hours came after that, not instead of it. If you do one thing Monday, pick a single HR process, ask who could tell you by Friday whether its data is clean and its rules are written down, and notice whether that person exists.
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.