| Last Week in HR AI: Issue #6 | Week of August 3, 2026 |
Knowledge capture, not automation, is where AI value shows up
Google's 15M-interaction study, the Fed's 490,000 earnings calls, and Lloyds pushing 80,000 people to AI fluency
|
1 arstechnica.com2 autonainews.com3 timesofindia.indiatimes.com4 shows.acast.com
Ikona's Take on the past week (August 3, 2026) in HR + AI
|
IO
|
Ian O'Keefe
Co-founder & CEO, Ikona Analytics
|
Two pieces of hard data landed this week that should change how you talk about AI in your next operating review. Google Research looked at 15 million Gemini interactions and found meaningful usage in only 21% of tracked work tasks, with AI touching most tasks in just 3% of occupations. The St. Louis Fed went through nearly 490,000 earnings calls and found that more than 90% of executive claims about AI productivity are forward-looking rather than realized. Aggregate productivity is flat. If you have been quietly wondering whether your peers are further along than you are, they are mostly further along in their slide decks.
The through-line we see this week is simple: the gap between promised AI value and measured AI value is closed by unglamorous data foundations work, not by better models or bolder announcements. Look at where value actually showed up. Indian manufacturers are pulling veteran operators' undocumented knowledge into knowledge graphs before those people retire, and one of them, Parle, tied a conversational tool to a near 10% cut in distributor inventory and a 5-6% sales lift. Lloyds built a joint control tower with the CTO, an ethics committee, and a specific target of moving 80,000 employees from AI literacy to AI fluency. None of that is a model story. It is a knowledge operations and governance story, which is the part most HR functions have never funded. And when four vendors launch "agent platforms" in a single month with four different definitions of the term, the differentiator is no longer guardrails; it is whether your organization has structured knowledge worth pointing an agent at.
The uncomfortable question for your team: if an agent went looking for how your HR processes actually run today, what would it find? In our engagements, the answer usually lives in the heads of about 50 SMEs and nowhere else, which is exactly why so many pilots stall before production. I would rather you spend this quarter capturing that than buying a platform to sit on top of nothing. If you are working through this right now, reply and tell me where you are getting stuck. I read every one.
| The six to read | The stories that matter for the Office of HR — with our take on each. |
Google Research released ATLAS, a study of 15 million anonymized Gemini interactions mapped against occupational task data. Meaningful Gemini usage appeared in only 21% of tracked work tasks, and AI touched most tasks in just 3% of occupations. Usage concentrated in drafting, information retrieval, and lower-expertise cognitive work, with the pattern pointing to complement rather than replacement. HR specialists ranked among the more AI-touched occupations in the data.
A July 2026 St. Louis Fed analysis of nearly 490,000 earnings calls found that more than nine in 10 executive claims about AI productivity describe expected rather than realized gains. Aggregate productivity remains flat, which the authors compare to electricity's long lag before factories were redesigned around it instead of simply re-powered. MIT research cited alongside it puts the share of firms capturing real generative AI value at roughly 5%. Most organizations stall between pilot and production and absorb a short-term productivity dip first.
JSW Group, Sun Pharma, and Parle Products are using AI to capture veteran employees' undocumented operational knowledge ahead of retirements. JSW connects equipment data with engineer expertise into knowledge graphs; Sun Pharma's assistant pulls from roughly 50 enterprise systems for about 20,000 managers. Parle's conversational tool, ChatG, is credited with cutting distributor inventory close to 10% while lifting sales 5-6%. Executives describe the approach as embedding AI into core operations rather than running isolated automation projects.
On the Digital HR Leaders podcast, Lloyds Banking Group Chief People and Places Officer Sharon Doherty walks through four years of reshaping the bank toward a fintech operating model. She describes a joint AI control tower run with the CTO, an ethics committee, and a cross-functional AI agent built with Microsoft. A stated goal is moving 80,000 employees from AI literacy to AI fluency. Doherty also makes the case that HR should own workplace and physical environment strategy.
Also covered by: shows.acast.com, shows.acast.com
OpenAI, Meta, NVIDIA with ServiceNow, and Google Cloud each launched enterprise AI agent platforms inside a single month, with four incompatible definitions of what an agent platform is: a distribution channel, a governed container, a managed service, and a data-gravity play. All four now lead their messaging with guardrails and audit capability. The analysis argues pricing models, not feature lists, reveal each vendor's actual strategy and intended buyer, since guardrails have become table stakes.
A 2026 study in Business & Information Systems Engineering ran a controlled experiment on whether relying on machine learning decision aids affects skill development. Participants who leaned on ML predictions during the task learned less, and their performance dropped sharply once the tool was removed. Trust in the system's outputs predicted how large that skill deficit became. The authors position the finding as a caution as generative AI use spreads across knowledge work.
Most people analytics functions sit on plenty of system data and almost no record of how the work actually gets done. That missing tacit knowledge layer is why generative AI projects in HR stall, and owning the context layer is the strategic claim available to analytics leaders now.
The most useful frame this week came from the Fed study's comparison to electricity: the gains did not arrive when factories swapped their power source, they arrived when factories were redesigned around it. That is the work in front of you, and it is slower and less quotable than an automation headline. But the study on machine learning decision aids is the caution I would keep on the wall: reliance without capability building leaves you with a sharp drop the moment the tool goes away. Leading a people function through this moment is genuinely hard, and the leaders making deliberate, foundational moves right now are the ones who will look like they read the tea leaves 18 months from now. Keep going.
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.