When a company builds artificial intelligence, its own workforce becomes the most visible proof. If the technology cannot improve how the builder operates, why should anyone else believe it works?
That reality places unusual pressure on AI companies. Their HR departments are not just managing people—they are also proving that the products their organisations sell can function in the messy reality of work. Hiring, onboarding and performance reviews become experiments that either validate or undermine the promise of AI-enabled HR.
The early results are revealing. Not because they demonstrate unambiguous success, but because they expose tensions that marketing materials often gloss over.
AI companies are discovering something uncomfortable: automating HR processes is relatively easy. Improving HR outcomes is far harder.
The internal proving ground
AI companies cannot credibly sell HR automation without demonstrating internal adoption. A consulting firm can plausibly advise on strategies it has not implemented. A technology company selling automation cannot.
This pressure turns internal HR operations into proving grounds for the technology before it reaches enterprise customers.
One of the most visible examples comes from Microsoft. The company empowered HR staff as “citizen developers,” allowing them to build AI tools that solved their own workflow problems. The results included an HR virtual agent that saved tens of thousands of hours in routine service work, along with AI integrations that reduced response times and sped up case resolution.
These gains are meaningful—but they occur in areas where AI performs best: structured, repetitive administrative tasks such as answering queries, retrieving policies and routing requests.
What is equally revealing is what followed. In 2025, Microsoft began encouraging AI adoption and signalling that AI usage would increasingly factor into employee productivity expectations. Staff were expected to demonstrate AI usage as part of their evaluations.
If the benefits were as obvious as productivity metrics suggest, mandates might not be necessary. Their presence hints at friction that efficiency gains alone do not explain.
The Anthropic contradiction
A different tension emerges at Anthropic, where internal research analysed a large internal dataset of employee interactions with its AI coding assistant.
The study found that employees were tackling more complex work when supported by AI. Tasks rated 3.2 on the company’s internal difficulty scale rose to 3.8, suggesting people were attempting more sophisticated problems with AI assistance. The company described the dynamic as 57 per cent augmentation and 43 per cent automation.
Yet Anthropic initially prohibited job candidates from using AI during applications, arguing it obscured genuine communication ability. The company reversed the policy months later, though AI remains banned in technical assessments—a partial retreat that suggests unresolved tension about when AI enhances work and when it distorts evaluation.
The reversal is more revealing than the original policy. The same organisation promoting AI as capability enhancer initially treated it as a distortion when evaluating candidates—then changed course when that position became commercially untenable.
The evolution points to an unresolved tension: when does AI enhance capability, and when does it obscure the qualities organisations are trying to assess? Anthropic’s policy shift suggests even AI companies building these tools have not settled the question.
The transfer problem
Even where AI clearly improves HR processes within technology firms, transferring those gains to other organisations is far from straightforward.
AI companies operate under conditions that are difficult to replicate elsewhere. Their employees tend to be technologically fluent, comfortable with rapid experimentation and supported by significant infrastructure. Cultural norms often prioritise speed over stability.
At OpenAI, Chief People Officer Arvind KC has been brought in as the company expands its people strategy alongside rapid AI-driven growth. Internal communication relies heavily on Slack rather than traditional email structures.
That environment bears little resemblance to the realities of large manufacturing firms, heavily regulated industries or organisations with distributed workforces and unionised labour.
Metrics showing faster HR case resolution in a technology company reflect those conditions. Whether similar gains translate to more traditional organisations remains an open question.
The metrics problem
Much of the public evidence about AI-enabled HR focuses on productivity metrics: hours saved, response times reduced, and administrative workloads automated.
These numbers are easy to measure and easy to market. They may also be misleading.
Faster case resolution does not necessarily produce better employee experiences. Automated onboarding may process more hires efficiently without improving how new employees integrate into teams. AI-generated job descriptions might increase consistency while removing the human judgement that attracts specific kinds of talent.
In other words, AI may be optimising HR for what is easy to measure rather than what actually matters.
The outcomes organisations ultimately care about—retention, leadership development, cultural cohesion and employee engagement—are slower, harder and far more complex to evaluate. Public reporting rarely includes them.
The risk is optimisation for proxies: making HR processes faster and cheaper without making organisations healthier or more effective.
Commercial incentives
AI companies also face commercial pressure to highlight the use cases that justify enterprise investment.
Recruitment workflows, onboarding documentation, policy management and internal communications all lend themselves to automation and clear ROI calculations. These are the areas where vendors focus their messaging.
What receives less attention are the parts of HR that rely on contextual judgement, relationship management and conflict navigation—activities that resist decomposition into tasks that machines can easily perform.
This imbalance is not dishonest. It reflects commercial reality. But it creates a skewed picture of what AI-enabled HR actually means.
Companies such as Google, Meta Platforms, and OpenAI increasingly promote AI applications for HR tasks—from drafting job descriptions to generating onboarding content. These internal uses double as product demonstrations for enterprise clients.
The laboratory therefore serves two purposes: operational experimentation and marketing validation.
What the early experiments suggest
Three years into the generative AI wave, some conclusions are becoming clearer.
AI is highly effective at automating structured administrative HR work. It can draft communications, retrieve policy information, generate standard documents and answer common employee queries.
These capabilities deliver measurable efficiency gains.
What remains uncertain is whether those efficiencies translate into better organisational outcomes. Faster HR processes do not necessarily create stronger cultures, better leadership pipelines or higher employee commitment.
The laboratory experiments underway inside AI firms will eventually clarify the relationship between the two. But the answers will emerge not from productivity dashboards alone, but from longer-term evidence—retention, leadership development, organisational resilience and workforce capability.
For now, the experiments show that AI can make HR faster and cheaper.
Whether it can make organisations better is something even the companies building the technology have yet to prove.