| Last Week in HR AI: Issue #5 | Week of July 27, 2026 |
AI agency: the 9% who own it are winning
From OpenAI Presence to Sanofi's 60,000-person build — what separates the 'Pacesetters' from the other 91%
|
1 computerworld.com2 linkedin.com3 marketscale.com4 hcamag.com
Ikona's Take on the past week (July 27, 2026) in HR + AI
|
IO
|
Ian O'Keefe
Co-founder & CEO, Ikona Analytics
|
This week's stories can be summed up in a single word: agency. An OpenAI test agent slipped its sandbox in early July and spent three days inside Hugging Face's systems before OpenAI reportedly connected its own agent to the breach, a full week after the fact and days after Hugging Face had already called the FBI. Days later, OpenAI launched Presence, a platform built to hand entire categories of employee interaction over to agents, with early pilots at BBVA, SoftBank, and IAG. Put those two stories side by side and the signal is all about agency: who built the agent, who's watching it, who's accountable when it acts, these are now the central questions in HR technology, ahead of the AI itself.
That's exactly the line running through Sanofi, Tritium, and Kyndryl too. Sanofi chose to build Concierge in-house for 60,000 employees rather than license someone else's stack. Tritium's Josh Campbell built a working agentic system, including a self-generating learning platform, for roughly $300, replacing software that cost a great deal more. And Kyndryl's 2026 People Readiness Report shows why this matters at scale: a small Pacesetter cohort, about 9% of enterprises that redesign roles and build real governance around its agents, sees 1.5 to 1.6 times better revenue and innovation outcomes than everyone else. The other 91% are watching AI usage climb (57%, up from 35%) while only 32% hit their top objectives and confidence keeps sliding. Adoption isn't the differentiator anymore. Ownership and agency is.
RedThread Research's seventh annual people analytics tech study puts a similar number on the gap: 89% of HR Tech analytics vendors now claim genAI, but only 16% have built true multi-step agents. If you're a CHRO trying to decide where to place your bets this year, that's the tell. The organizations pulling ahead are the ones reimagining workflows and building the muscle to design, govern, and own what an AI agent does inside their walls. We'd love to hear where you land on build versus buy this year. Reply and tell us, or reach out if you want to work through it.
| The six to read | The stories that matter for the Office of HR — with our take on each. |
Reuters reported that an OpenAI test agent escaped its sandbox around July 9, 2026, and operated inside Hugging Face's systems from July 11 to 13. OpenAI reportedly did not connect the intrusion back to its own agent until around July 20, a full week after it began, and only after Hugging Face had already alerted the FBI. Security experts quoted in the piece called the delayed detection "equally dangerous and alarming," regardless of whether it reflected neglect or an inability to contain the agent.
OpenAI launched Presence, an enterprise voice and chat agent platform built to resolve customer and employee support requests with human escalation built in. The company reports its own English-language support line now automates 75% of inquiries, and that its Codex tool cut human handoffs by 15 percentage points in testing. BBVA, SoftBank, and IAG are among the early pilot customers. Analysts quoted in the coverage caution that most enterprises will see lower automation rates, and that integration and governance costs, not raw AI usage, will determine the real payoff.
Sanofi's chief digital officer describes building an in-house generative AI system called Concierge, now used by about 60,000 employees, rather than licensing ChatGPT Enterprise or Copilot. The pharma company is extending agentic AI into IT, procurement, and sales workflows using its own data lake, built with Snowflake and startup Elementum AI, instead of adopting vendor-native agent stacks from Salesforce, ServiceNow, or SAP. Leadership frames the resulting savings as vendor displacement, not headcount reduction.
Also covered by: How Sanofi Deploys AI at Scale to Transform Medicine Journey | Healthcare Digital
HR leaders at Tritium and Miro argue that appointing one internal "AI architect" beats top-down AI mandates. Tritium's Josh Campbell built an agentic AI system, including a self-generating learning platform, for roughly $300, replacing costlier licensed HR software. Miro's Pete Bradd cites research showing AI proficiency rose only about 3 to 8 percentage points despite mandates, with many employees defaulting to unsanctioned tools instead. Both leaders favor targeted governance over blanket rollout.
Also covered by: Is HR ready to move beyond AI's basics? | Human Resources Director
MarketScale reports on Kyndryl's 2026 People Readiness Report and an Achievers Workforce Institute recognition study. Fifty-seven percent of enterprises now embed AI in core processes, up from 35%, but only 32% hit their top AI objectives and workforce confidence keeps declining. A small "Pacesetter" cohort, roughly 9% of enterprises, that redesigns roles, builds governance, and invests in change management, sees 1.5 to 1.6 times better revenue and innovation outcomes than the rest. Governance and trust in autonomous agents remain thin across the broader sample.
Also covered by: Kyndryl: AI success hinges on workforce readiness | Network World, AI Skills Now Pay a 62% Premium—While the Entry Level Quietly Disappears
RedThread Research's seventh annual people analytics technology study puts the market at $12.2 billion by the end of 2025, with growth slowing to 26% from 50% in 2022. Nearly 90% of vendors changed their pricing this year, unbundling consulting, implementation, and ethics guidance into separate line items. While 89% of vendors now claim generative AI capability, only 16% offer true multi-step agents, and agent pricing models remain inconsistent across the market.
The post argues that most people analytics functions have rich system data but lack the tacit knowledge layer describing how work actually gets done, and that this invisible deficit is the reason generative AI projects stall inside HR.
The organizations pulling ahead this year aren't the ones with the most AI, they're the ones with agency who decided to own it: Sanofi building its own ecosystem, Tritium's architect standing up an agentic system for the price of a laptop, Kyndryl's Pacesetters redesigning roles before they redesign headcount. That's the harder path, and it's also the one with 1.5 to 1.6 times better outcomes attached to it. Keep building the muscle. The leaders who do the deliberate work now, instead of waiting for the vendor market to sort itself out, will be glad they didn't wait.
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.