Where AI Stands in People & Talent
No corporate function has absorbed more AI than HR, and none has less to show for it. Applicant tracking systems reach 98% of the Fortune 500; 83% of employers run automated resume screening; 87% of companies use AI somewhere in hiring; 39% have deployed it formally across HR functions and 62% use it informally somewhere. Against that, the same surveys report that 88% of HR leaders have realised no significant business value, that 56% do not formally measure whether they have, and that only 6.6% of enterprises have integrated AI end-to-end across the hiring process at all. The gap is not a measurement artefact. Asked to name their biggest obstacle, 56.7% of HR leaders cite integration with existing systems — ahead of bias, privacy, cost, or model quality. This is a domain where the technology question closed some time ago and the organisational question has barely opened.
The result is a landscape that has stopped moving. Every practice we track in People & Talent is holding its position, and with one exception every trajectory is flat or falling. Resume screening and candidate sourcing are genuinely established — ubiquitous, commoditised, and priced accordingly. A middle band of good-practice capabilities (onboarding automation, HR policy chatbots, engagement analytics, adaptive assessment, personalised learning, skills mapping, attrition prediction, workplace analytics) has mature generally available tooling and named production deployments but sits on an adoption plateau that has now persisted for four consecutive scan cycles. The leading-edge band — structured assessment scoring, compensation benchmarking, performance review synthesis, organisational network analysis, workforce planning — is where vendors have shipped the most sophisticated agentic capability and where organisational adoption is thinnest. DEI analytics is the only practice actively declining, and it is declining for reasons that have nothing to do with whether the tools work.
Underneath the plateau, the domain has split into two populations. Enterprises with centralised talent data, a dedicated people-analytics function, and a compliance department extract returns that are large and verifiable: Chipotle cut time-to-hire 75% (twelve days to under four) while lifting application completion from 50% to 85%; Unilever saved 50,000 recruiter hours and £1M a year; Workday's own talent team recovered 24,000 hours in nine months and redeployed 40% of its capacity; IBM's AskHR contains 94% of queries across 11.5 million interactions a year; Visier's Paycor partnership runs attrition modelling across 2.1 million employees. Everyone else has bought capability they cannot wire in. Only 31.6% of US business managers at smaller employers use AI in hiring at all, and among those only 13.8% report significant improvement. The bifurcation is not about model access — it is about whether the organisation has clean data, defined decision rights, and managers willing to use the thing.
What's New, 2026-07-28 to 2026-08-11
The sharpest new evidence of the fortnight converges on a single distinction: AI is good at coordination and bad at judgment, and organisations are deploying it in inverse proportion. PYX Labs published a peer-reviewed benchmark of seven frontier models across 84 employee-listening tasks and found scores of 64–82% on categorisation and classification but as low as 33% on synthesis — the interpretive work of reading feedback and forming a view. In the same window, a Resume Builder survey found 60% of managers already use AI in decisions about raises and layoffs, with fewer than 20% having received any formal training. Culture Amp's study of 264 HR professionals (6 August) put the corollary in adoption terms: 77% say they are bought in on AI, only 24% are comfortable with agentic deployment, and organisations that report transformation see two-to-three times better outcomes specifically when automation targets coordination rather than judgment-making. Three independent sources, one finding. It is the clearest guidance this domain has produced in months, and it cuts against the direction most agentic HR roadmaps are pointed.
The second cluster is about the labour market itself, and it is uncomfortable. A Gartner CHRO survey published 9 August found 22% of organisations have frozen entry-level hiring on the grounds of AI automation, even though only 5% report any measurable employment impact from AI — a decision being made on narrative rather than evidence, with Stanford data showing a 16% relative employment decline for 22-to-25-year-olds in AI-exposed roles as the consequence. Circana's forecast (10 August) extends AI-driven hiring disruption through 2027 with no inflection until 2028, describing a low-hire, low-fire environment rather than a displacement event; Futurum's assessment three and a half years after ChatGPT's release finds no economy-wide job displacement and roughly 1% measured productivity gain. Kelly Services data adds that 55% of leaders who cut jobs for AI now regret it. Meanwhile Gallup's annual benchmark (7 August) recorded global engagement at 20%, a twenty-year low, with manager engagement down nine points since 2022 and an estimated $10 trillion annual cost, and a Deloitte/Mercer synthesis showed employee thriving collapsing from 66% to 44% while fear of obsolescence rose from 28% to 40%. Elsewhere: compensation governance took two hits — the Capitol Forum opened an investigation into whether Pave's 9,000-company salary data pool enables wage coordination among competitors, while separate survey data found 84% of compensation professionals still run high-stakes pay analysis through general-purpose ChatGPT rather than a compensation-specific tool. Enforcement continued to land, with a $3.2M DOJ settlement against OpenAI and Statsig over discriminatory hiring practices and a $15M punitive jury verdict against Nike on gender pay equity that a rigorous audit trail did not prevent. Nothing changed tier or trajectory. In a domain this saturated, stability is the story.
Key Tensions
Automating coordination pays; automating judgment does not — and organisations are doing the reverse. The evidence is now triangulated: frontier models score 64–82% on categorising employee feedback and 33% on synthesising it, yet 60% of managers use AI in pay and exit decisions with under 20% trained. Culture Amp's transformation cohort achieves two-to-three times better outcomes precisely by keeping automation on the coordination side of the line. The vendors selling into this domain are shipping agentic capability aimed squarely at the judgment side; Josh Bersin counted 24 discrete agent use cases across the major talent acquisition platforms in mid-July alone. beqom's own guidance recommends against end-to-end autonomous agents in compensation, and 15 CHROs surveyed on their AI guardrails were unanimous that pay decisions require human authority — a rare case of vendors and buyers agreeing that the roadmap is ahead of the use case.
Human oversight is the control everybody claims and nobody has validated. Regulators from Colorado to Brussels have made "meaningful human review" the load-bearing safeguard, but the empirical record is bad. A University of Washington study of 528 reviewers across 1,526 screening scenarios found they mirrored biased AI recommendations up to 90% of the time, including when they rated the output as poor; FAccT research finds AI recommendations cut recruiter deliberation time by 55%. Non-determinism compounds it: running the same screening tool twice on an identical candidate pool produced only 14% shortlist overlap. Meanwhile 78% of organisations deploying hiring AI lack a structured bias assessment framework and fewer than 25% have documented impact-testing procedures. The control that regulation depends on is, on current evidence, largely ceremonial.
Two regulatory regimes are now moving in opposite directions, and multinationals are caught between them. Europe is building: the EU AI Act's Article 50 transparency duties and the emotion-recognition prohibition took effect on 2 August 2026, the Pay Transparency Directive's reporting obligations bind employers above 250 staff, and high-risk documentation follows in December 2027 with penalties up to 7% of global revenue. The United States is dismantling: the EEOC has proposed eliminating the EEO-1 through EEO-6 reporting framework in place since 1966, DEI-function AI adoption has fallen to 2% or below against 39% across HR overall, Fortune 500 public DEI disclosure dropped 65% year-on-year, and NALP's 35-year diversity benchmark lost 30% of its participants. Yet the litigation risk in the US is rising, not falling — Mobley v. Workday covers 1.1 billion rejected applications, Kistler v. Eightfold extends FCRA consumer-protection liability to candidate scoring across a billion worker profiles, and the Nike verdict shows a clean audit is not a defence when a jury sees it differently. Organisations are losing their measurement baseline exactly as their exposure grows.
Sourcing got an order of magnitude cheaper; hiring did not get better. This is the most under-appreciated finding of the cycle. An executive search firm's analysis (3 August) documents that while the cost of finding candidates has fallen roughly tenfold, the senior hire failure rate remains around 30% — unchanged since 2018 — because the bottleneck was never sourcing but role scoping, fit diagnosis, and judgment. The channel data says the same thing from the other end: platform telemetry across 13.2 million LinkedIn connection requests shows acceptance rates holding at 28% while reply conversion collapsed from 32.2% to 22%, and analysis of 844,000 recruiting sequences finds 64.9% of replies arrive after the first message, meaning single-touch automation captures barely a third of interested candidates. Signal-based outreach (funding events, leadership changes, headcount velocity) achieves 3–7% reply rates against 0.3–0.8% for generic cold contact. Precision beats volume by roughly ten to one, and the automation industry has spent three years optimising volume.
Managers, not models, are the variance. Every practice in this domain that stalls, stalls at the manager. Gallup's study of 43,000 US workers found engagement at 48% where managers actively support AI use against 30% where they do not, rising to 53% when frequent use, a clear organisational plan, and manager coaching all align. Fuel50 platform data shows manager active use — not endorsement — drives three-to-five-times faster adoption of skills systems, and only 34% of organisations achieve strong participation. In internal mobility, 46% of managers actively resist internal moves despite internal hires costing 50–60% less and staying 60% longer. Burning Glass and Harvard found that among firms that dropped degree requirements, only 37% actually changed who they hired. The tooling assumes a manager who will use it; the incentive structure produces a manager who will not.
The value case for headcount reduction has quietly collapsed, and workforce planning is rebuilding around that. Fifty-five percent of business leaders who cut jobs on the strength of AI forecasts now regret the decision, with Ford, Commonwealth Bank, and IBM all rehiring within six months; Stanford research finds augmentation strategies outperform replacement. That matters because headcount reduction was the assumed ROI mechanism for the entire workforce planning category, which carries first-year implementation costs of $500K–$2.5M and reaches fewer than 15% of enterprises. With only 23% of executives reporting their workforce is ready for AI (down from 29%), 75% of organisations holding no documented AI workforce plan, and only 20.7% of enterprises with skills-inventory coverage above 75%, the practice is being asked to justify itself on a different basis than the one it was sold on.
Top 10 Evidence Items
Six in Ten Managers Use AI to Decide Raises and Layoffs. A New I-O Benchmark Shows It's Weakest Exactly There. (research-paper) — The clearest single piece of evidence behind the domain's central finding this cycle: the PYX Labs benchmark scores frontier models 64–82% on categorising employee feedback but as low as 33% on synthesis, while 60% of managers already use AI to decide raises and layoffs with under 20% trained — automation deployed exactly backwards from where it is reliable. https://scovai.com/blog/six-in-ten-managers-use-ai-to-decide-raises-and-layoffs-a-new-i-o-benchmark-show/
The Manager Is the Missing Multiplier: Employees Adopt New Skills 3-5x Faster When the Boss Uses the Tool (adoption-metric) — Confirms the domain's most persistent bottleneck is manager behaviour, not model capability: skills-system adoption runs three-to-five times faster when a manager actively uses the tool, explaining why only 34% of organisations achieve strong participation despite mature, generally-available tooling. https://blog.theinterviewguys.com/the-manager-is-the-missing-multiplier-employees-adopt-new-skills-3-5x/
The Cost Paradox: Why Companies Are Freezing Entry-Level Hiring Before AI Actually Works (adoption-metric) — Documents the labour-market story behind Gartner's finding that 22% of organisations have frozen entry-level hiring against only 5% reporting any measurable AI employment impact — a decision being made on narrative rather than evidence. https://finance.yahoo.com/technology/ai/articles/cost-paradox-why-companies-freezing-131317693.html
AI's Workforce Shock Is Unlikely To Ease In Near Term: Circana (industry-report) — The forecasting evidence for a low-hire, low-fire labour market rather than a displacement event, extending disruption through 2027 with no inflection expected until 2028. https://www.crn.com/events/2026/ai-s-workforce-shock-is-unlikely-to-ease-in-near-term-circana
Gallup 2026 Workplace Report: HR, IT & Ops Insights (industry-report) — The engagement collapse behind the entire market case for engagement analytics: global engagement at 20%, a twenty-year low, with manager engagement down nine points since 2022 and an estimated $10 trillion annual cost — the crisis the practice exists to solve is deepening even as adoption plateaus. https://www.mangoapps.com/articles/gallup-2026-state-of-the-global-workplace
Pave Salary Data Access Raises Concerns Over Collusion Risk (opinion) — The compensation-benchmarking practice's governance problem made concrete: a formal Capitol Forum investigation into whether Pave's 9,000-company salary data pool enables wage coordination among competing employers. https://thecapitolforum.com/pave-compensation-benchmarking-service-provides-customers-access-to-sensitive-non-public-salary-data-raising-questions-about-potential-collusive-effect/
OpenAI pays $3.2m to settle claims it discriminated against US workers (news-coverage) — Shows US enforcement risk rising even as DEI reporting infrastructure is dismantled — a federal settlement landing against a frontier AI lab itself, over discriminatory hiring practices via the OpenAI/Statsig hiring tool. https://www.theguardian.com/us-news/2026/aug/04/openai-worker-discrimination-claims-settlement
Pay Equity Audits: Lessons from Nike Case (opinion) — The evidentiary basis for the summary's warning that a documented audit trail is not a legal defence: a $15M punitive jury verdict landed against Nike on gender pay equity despite rigorous compensation governance processes. https://morganhr.com/blog/pay-equity-audits-after-the-nike-verdict-managing-litigation-risk-you-cannot-see-on-a-spreadsheet/
AI Has Made Sourcing Cheap. The Hires Are Still Failing. (opinion) — The sharpest articulation of the domain's most under-appreciated finding: candidate-sourcing costs have fallen roughly tenfold while the senior-hire failure rate has held near 30% since 2018, because the bottleneck was always role scoping and judgment, not sourcing volume. https://www.jenningsexec.com/insights/ai-has-made-sourcing-cheap-hires-are-still-failing-at-the-same-rate
We applied to be a Kmart holiday casual. An AI bot did the interview (case-study) — A ground-level, candidate-facing illustration of ubiquitous-but-ungoverned hiring automation: a retail seasonal application processed end-to-end by an AI interview bot, the sharp end of the statistic that 87% of employers now use AI somewhere in hiring. https://www.smh.com.au/business/workplace/we-applied-to-a-job-at-kmart-an-ai-bot-did-the-interview-20260805-p60lof.html