Skip to main content
All issues
Share:
Last Week in HR AI: Issue #21 Week of November 16, 2026

Approve an HR build before mapping the work and you own the mess

One HR team's early free-for-all building produced duplicate tools, so guardrails are being added now.

This week in 60 seconds 6 signals
1 Remote's people team built its own applicant tracking system with Claude; 24,000+ applications handled since October itbrew.com
2 US Navy personnel runs 164 separate HR systems with no shared data architecture exa.ai
3 AT&T: nearly 500 agentic workflows in production, set to triple next year, with fivefold returns within a year SiliconANGLE
4 Orange Business hit its €50M AI value target for 2025 and raised it to €70M, with 100,000+ active users telecomgazette.com
5 McKinsey: about 40% of large organizations used AI agents, up from 27% the prior year theguardian.com
6 OECD 2026: up to a third of highly skilled workers in some countries rarely apply their skills Forbes
1641
Navy HR systems with no shared architecture
24,0002
Applications through Remote's homegrown ATS
€70 million3
Orange Business's raised AI value target
40%4
Large organizations using AI agents

1 exa.ai2 itbrew.com3 telecomgazette.com4 theguardian.com

The Week in Review

Ikona's Take on the past week (November 16, 2026) in HR + AI

IO
Ian O'Keefe
Co-founder & CEO, Ikona Analytics

The advantage this week went to organizations that own their own plumbing, and the question for a CHRO has moved from which vendor to buy to what your function can build and run itself. Remote's people team let recruiting and engagement contracts lapse and built its own applicant tracking system, mostly with non-engineers. It has handled more than 24,000 applications. AT&T has nearly 500 agentic workflows in production.

Building is now cheap; knowing what to build is not. The Navy's top personnel official inherited 164 HR systems with no shared data architecture, and he says earlier data-lake efforts failed because they copied old processes. AT&T calls consolidating data left by years of acquisitions a prerequisite, and its foundry maps a process before an agent runs it. Remote's early free-for-all produced duplication. Ownership without a map of how your function actually works reproduces the mess faster, on infrastructure you now have to maintain yourself.

The skills research in this issue makes the same point about people: capability only counts when it is used where the business needs it. Systems work the same way. Before you approve a build, ask whether your team could write down how hiring, pay, or succession actually runs today, well enough for an agent to follow it. If the answer is no, that is your first project. Reply and tell me where your team is building, or reach out if you want to dig in. We'd welcome the conversation.

Have a reaction to this? Contact us →

The six to read The stories that matter for the Office of HR — with our take on each.
01 itbrew.com
Remote’s HR team ditched external vendors and is building their own HR systems with AI

Barbara Matthews, chief people officer at Remote, allowed contracts for outside recruiting and engagement software to expire and asked her own team to replace them. Most of the builders are not engineers; they work in Claude and Claude Code. Their applicant tracking system went live in October and has processed more than 24,000 applications and candidates, including records migrated from the old tool. A learning management system, a 360 feedback tool, and engagement surveys came next, and after early duplication the team is now putting guardrails in place.

Ikona's take
This is the opening argument of the week: a people team without engineers replaced vendor software and is running it at real volume. When building costs this little, the scarce input becomes a precise account of how recruiting and feedback actually work inside your company. Note the sequence, too: the guardrails arrived after the duplication, and a map drawn up front lets you skip that step.

Read the full article →

02 exa.ai
Can AI Help the U.S Navy Decide Who Gets Promoted? Assistant Secretary of Navy Ben Kohlmann (M&RA)

Assistant Secretary of the Navy Ben Kohlmann, the service's top personnel official, says he arrived to find 164 separate HR systems with no shared data architecture, which left basic workforce questions unanswered. He supports consolidating that data into a single total-force view covering military personnel, civilians, reservists, and contractors. In his account, earlier data-lake projects failed because they reproduced legacy processes, and he sees AI making it more feasible for government to build in-house. Despite the headline, the excerpt does not reach promotion decisions.

Ikona's take
This is the flip side of the AT&T story: what you get when you build before you map. Kohlmann's verdict on the old data lakes is the warning for every CHRO inspired by Remote, because copying a broken process onto new infrastructure gives you the same broken process with a bigger maintenance bill. Before consolidating anything, document which questions your systems were supposed to answer and who actually answers them today.

Read the full article →

03 SiliconANGLE
AT&T uses agentic AI to reinvent business processes from the ground up - SiliconANGLE

SiliconANGLE reports that AT&T runs close to 500 agentic workflows in production and expects to triple that count next year, with the aim of redesigning entire processes instead of automating isolated tasks. A central foundry maps each process, stores the business logic in shared memory, and sends customer-facing workflows through security and compliance review. Agents operate on their own only after their accuracy reaches the mid-to-high 90s. An AT&T leader cites fivefold returns inside a year and names consolidation of data inherited from years of acquisitions as a prerequisite.

Ikona's take
AT&T shows what the Remote model looks like at scale, and the order of operations is the lesson: the foundry maps the process and writes down the business logic before any agent runs it. That shared memory is the real asset, and it is the context most HR functions have never captured, a gap we dig into in Context is all you need. If your team cannot describe a process precisely enough to measure an agent's accuracy against it, you are not ready to set a threshold.

Read the full article →

From Ikona
Your build is only as good as your process map

We diagnose how work and knowledge actually flow through your HR function and capture the judgment that lives in people's heads. Then we structure it into a map an agent, or your own build team, can follow.

Talk to us about yours →
04 telecomgazette.com
Orange banks hundreds of millions in AI value — without layoffs — Telecom Gazette

Orange Business Services met its €50 million AI value goal for 2025 and has raised the target to €70 million, according to its chief data and AI officer. The unit counts 30 use cases in production, 20 more being industrialized, and over 100,000 active users. He attributes the gains to speed rather than job cuts, pointing out that Orange's headcount had started falling before ChatGPT. Orange runs a mix of commercial and open-weight models, keeps its agents portable, and he names a transformation that falls behind the technology as the main risk.

Ikona's take
Portable agents are the ownership argument in another form: Orange will not let a single model vendor hold its plumbing. The more useful line for a CHRO is the warning about transformation lagging the technology, because that lag sits in HR, where roles and processes get redesigned. Measure value in euros and speed the way Orange does, and your business case stops depending on a headcount number.

Read the full article →

05 theguardian.com
What is it like to work when your co-workers are AI agents? | Technology

The Guardian profiles workplaces where AI agents carry much of the load, including the startup Anything, where about 15 people work with roughly 150 agents covering sales, marketing, and coding. The feature describes the friction: customers and colleagues who want a human, roles being displaced, and employees who spend their time supervising agents that make mistakes. It cites McKinsey data showing about 40% of large organizations used agents between May 2025 and June 2026, compared with 27% the year before.

Ikona's take
When employees become managers of error-prone agents, the job design lands on HR, and it is the people version of this week's argument. Someone has to define what the agent owns, what the human checks, and when a customer gets a person. Write that down before you deploy, or your strongest people become full-time bot supervisors.

Also covered by: wired.com

Read the full article →

06 Forbes
Talent Management Was Built For Stability; Work No Longer Is

A Forbes contributor contends that hiring, reskilling, succession, and learning programs were designed for stable work and fall short now. Drawing on 2026 OECD research, the article reports a weak connection between how proficient workers are and whether they use those skills, with as many as a third of highly skilled workers in some countries seldom applying them. It calls on HR to test whether skills turn into capability deployed where the business needs it, since shortages and underuse exist side by side.

Ikona's take
This closes the loop: a skill nobody uses is the people equivalent of a system nobody mapped, and both look like assets on paper. Pick one process, hiring, pay, or succession, and ask whether your team could write down how it runs today well enough for an agent to follow it, including where skilled people are actually deployed. If they cannot, that document is your first build, and every build after it gets cheaper.

Read the full article →

From Ikona
Context is all you need

Every meeting is a way of getting context into a room, and software now makes calls that used to pass through that room, in HR with almost none of the context a person would bring. The post argues that closing the gap is the next thing HR owns, and that no function has further to go.

Ikona's take
We're resurfacing Context is all you need because every story this week turns on context: AT&T's shared memory of business logic, the Navy's 164 disconnected systems, and Remote's early duplication. Building is cheap now; the context an agent needs to make a good call is what your function has to capture first.
Before you go

Hold two numbers from this week side by side: Remote's in-house applicant tracking system, which has already handled more than 24,000 applications, and the Navy's 164 HR systems with no shared data architecture. The second is what the first can turn into when nobody draws the map, and Remote saw the early signs in its own duplication, which is why guardrails are going in now. Orange's data chief named the main risk as a transformation that lags the technology. For a people team that can ship its own tools, that lag is the distance between how fast you can build and how clearly you can describe what you are building for. Close that distance first, and AT&T's nearly 500 agentic workflows start to read less like a telecom headline and more like a plan your function could write.

Working through one of these challenges?

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 →
Share:

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.