Second edition · July 2026

Mastering AI for HR. What changed since 2024.

In October 2024 we interviewed 28 HR leaders and published a CPO's playbook for getting value from AI. Twenty-one months later, adoption is much higher but reported business value is not. Regulation and litigation have moved in different directions, workforce reductions have produced mixed results, and agents have become a real category of capacity. The 2026 edition tests the original eight chapters against published evidence from 2025 and 2026 and adds two chapters on agents and candidate fraud.

Nearly 300 heads of people, CHROs, and founders read the first edition. This one synthesizes 30+ published sources, cited in line: SHRM, Gartner, McKinsey, BCG, MIT, Stanford, Pew, ILO, Challenger Gray, court filings, and vendor disclosures.

The scoreboard

Ten numbers that tell the story. Adoption is no longer the constraint. Value, governance, and work design are. Where a figure is vendor-reported or definitionally contested, the report says so in line.

43%

of organizations now use AI for HR tasks, up from the edition's 12% anchor in 2024.

SHRM, 2025
88%

of HR leaders say they have not realized significant business value from AI tools.

Gartner, Oct 2025
14%

actually use the AI inside their core HR system. 57% say their HRIS has none.

SHRM, Dec 2025
95%

of enterprise GenAI pilots show no measurable P&L return.

MIT NANDA, Jul 2025
70%

of AI value comes from process and work redesign. The algorithms are 10%.

BCG, Sep 2025
~73%

of AI-justified layoffs report rehiring costs that met or exceeded the savings.

Careerminds, Feb 2026
13-16%

relative employment decline for workers aged 22-25 in the most AI-exposed occupations.

Stanford Digital Economy Lab, 2025
~1.1B

rejected applications behind the Mobley v. Workday collective. Enforcement came from plaintiffs rather than regulators.

Court filings, 2026
130+

digital employees at BNY with their own logins, email addresses, and human managers.

CNBC, Feb 2026
92%

of large-enterprise security leaders lack full visibility into their AI agent identities.

CISO survey via Strata, 2026
Adoption and reported value have moved at different rates.
Using AI for HR · 2024, the edition's anchor
12%
Using AI for HR · 2026
43%
Realizing significant business value · 2026
12%
Anchor: McKinsey 2024 · SHRM 2025 · Gartner Oct 2025 (value = complement of the 88% reporting none)
Every chapter that follows is about that gap: where it hides, what it costs, and the specific move that closes it. A number you cannot interrogate is not evidence.

The ten chapters

Eight chapters from 2024, re-scored against what actually happened, plus two chapters that did not exist the first time. Each ends in the specific move that closes the gap.

01

Start with the AI already in your HR stack

The vendors shipped agents into HR systems, while reported use inside the core system remains low. Inventory what you already own before buying another line item.

02

Fluency matters, but managers are the constraint

Five or more hours of training is associated with regular use. Only 8% of HR leaders believe their managers can use AI effectively.

03

Keep consequential people decisions human

66% of surveyed managers used AI to help decide layoffs, and roughly one in five often or always let it decide without human input.

04

Audit for litigation as well as regulation

Some regulatory obligations were delayed or weakened while private litigation advanced. Build an audit record that can withstand either route.

05

Work redesign determines whether job cuts hold

Companies are reporting substantial rehiring after AI-driven cuts. The relevant distinction is whether the work was redesigned before the reduction.

06

Automation without redesign still fails

95% of enterprise pilots show no measurable P&L impact in one widely cited study. The strongest evidence points to process and work redesign.

07

HR's role in AI strategy is still contested

52% of organizations exclude HR from AI strategy. A work inventory and decision-rights table make the case for HR's operating role.

08

Consent is now an operating requirement

Worker trust is low and labor agreements increasingly address AI. Write the consent architecture before deployment.

09

Planning a workforce that includes agents

Some companies now treat software agents as a category of capacity. Workforce plans need owners, unit costs, access rules, and review cadences for them.

10

Identity verification belongs in the hiring funnel

Gartner projects one in four candidate profiles will be fake by 2028; 41% of security leaders say their organization has already onboarded a fraudulent candidate.

Keeping score: what we got wrong

Here is where the first edition missed, named plainly.

Wrong mechanism, right conclusion

We expected regulators to force audit discipline. The EEOC withdrew, the EU deferred, Colorado gutted its act, and the enforcement wave arrived through the courts instead. Audit regardless; the reader changed, the requirement did not.

We underweighted the entry level

The job impact concentrated on the youngest workers faster than we projected: a 13-16% relative employment decline at ages 22-25, and new grads down to roughly 7% of Big Tech hiring. The adjustment is being carried by people with the least institutional power.

We did not say the word "agents"

Nobody gets that one back. The 2024 edition treated AI as a capability inside tools. Twenty-one months later it is a workforce category with logins, managers, and a spend line growing 140% year over year.

Get the 2026 edition

Fourteen pages, ten chapters, one honest scorecard. Read the 2026 edition.

How we scored ourselves. The 2024 edition drew on 28 practitioner interviews. This edition trades that intimacy for verifiability: 30+ published sources from 2025-2026, cited in line, including SHRM, Gartner, McKinsey, BCG, MIT NANDA, Stanford, Pew, ILO, Challenger Gray, court filings, and vendor disclosures, plus field observations from TL Partners engagements. Conflicting figures are presented with their definitions throughout, because a number you cannot interrogate is not evidence.

Running an AI reset, or backing companies that are?

The report is the thinking. The work is the reset: the inventory that finds the 15-20% of surviving work with no remaining consumer, the agent boundary published per function, and the fluency program behind the new model. If that's the moment your company or your portfolio is in, 25 minutes is the right amount of time to find out. thomas@tlpartners.biz