AI agents in HR: Work without Humans. It's coming by the end of the year. Really?
- Jean-Baptiste Audrerie
- Feb 26
- 9 min read
Updated: Feb 27
Work without humans. That's for the end of December 2026. The Californian techno-messianics are telling us so. Really?
Within 12 to 18 months, white-collar jobs will be completely automated. This is what Dario Amodei, CEO of Anthropic, told us at the Davos forum, as did Mustafa Suleyman, CEO of Microsoft AI, and Sam Altman, who warned us at the World AI Summit about the destruction of all jobs. Yes, ALL of them. Even his own, surely!
Autonomous AI agents will have replaced engineers, financial advisors, business analysts, and HR professionals. The AI apocalypse is coming by the end of 2026. They want to impress us.
The signals are certainly multiplying, but they are also misleading, designed for capital markets to finance servers and R&D with billions of dollars. The reality: automated tasks do not replace human labor, unless the labor itself is an automatable task.
Facts about AI agents in HR and at work:
Claude Code and Claude Cobol from Anthropic improve programmers' coding workload by 70 to 90%. Vibe Coding is disrupting the entire SaaS software industry value chain. All software vendors have slowed hiring and are training their programmers. The profession has changed dramatically in less than 18 months. All it takes is for Claude to announce that he can " vibe code " in Cobol (a language from the 1970s) and fix any type of code for IBM's stock price to plummet.
Claude Excel automates data cleaning and categorization (though still with many errors), as well as the production of analytics and dashboards, and even better, the annotated analysis of data. If you have data, your life will change.
Claude PowerPoint transforms a text into a PPT report.
OpenClaw (acquired by OpenAI) is launching a swarm of autonomous AI agents on your computer, and a two-hour workweek is being announced for all knowledge workers. The agents develop responses and code amongst themselves. And soon they'll be joining Teams for a virtual coffee break without your consent?
Claude Cowork , which I'm testing for a few days after some difficulty installing it on Windows 11, lets you create workflows and build a plethora of AI agents (called Skills). Connected to your computer and your work applications (Google Drive, Gmail, Outlook, Asana, Notion, etc.), it becomes a small orchestrating brain. You have to tell it what to do. It's not autonomous enough to impose itself and take over your work methods, at least not yet, but that shouldn't be long in coming, given how well the AI can "mimic" our behaviors and language. Fantastic. Impressive. But these are tasks, not a job, that are being automated. They want to impress us. We're dazzled, just as our ancestors were less than 130 years ago when electricity, light bulbs, and cars were introduced. Horses disappeared from our streets. There has never been so much travel.
If you spend all day coding, using Excel, PowerPoint, entering data, or answering customer calls, then your job has indeed just changed and you are at very high risk of being automated.
If you produce fairly complex and relatively standardized reports, analyses and calculations using a precise method, the output is indeed fast and increasingly well controlled.
" If the value of knowledge tends towards zero, the value of knowledge orchestration increases ." - Jean-Baptiste Audrerie
Thinking about the "First and Last Human Mile of AI".
The “ First and Last Human Mile of AI ” is an idea we formulated in the 1st AI in HR Barometer in Canada ( download available here ).
The concept is as follows:
The first kilometer is about intention and the overall contextual window.
While AI is constantly improving the contextual windows with which it interacts and personalizing its outputs, the initial vision and intentions remain the domain of entrepreneurs, engineers, innovators, and tech enthusiasts. Teams and AI agents must be managed and assembled according to a unique approach for each (know-how, expertise, patents, data).
The glue that binds the tasks that make up a job and the jobs that make up a company is highly intangible, implicit, and intentional. A white-collar professional spends more than 50% of their time performing complex cognitive tasks.
The rest? : relational, political, emotional, creative, collaborative, social, cultural, stakeholder alignment, engagement, psychological and change.
The last kilometer is where the decision-making happens.
We are entering a highly sensitive and risky area, with legal, societal, and moral implications. Governing decisions with semi-autonomous and autonomous AI is a real challenge, apart from certain purely transactional tasks that are very repeatable (True or False, Go/No Go) and devoid of moral judgment.
" The knowledge worker is fascinated and amazed by the ingenuity of synthetic thinking and its computing power to mimic human reasoning . So he consoles himself as best he can by highlighting the technical limitations of these dazzling new capabilities." Jean-Baptiste Audrerie
The perspective of an organizational psychologist and work analyst like myself.
Between 80 and 130 major tasks comprise the work of an HR director, an engineer, or a tax specialist. Most studies (and our common sense) converge on several hundred micro-tasks. Publishing an article on this blog or LinkedIn requires at least ten coordinated tasks, and that's not the core of my work as a consultant. It's these micro-tasks, or discrete, often implicit tasks, that AI agents can handle if we're willing to formalize, automate, and combine them. Quite a challenge!
Only 30 to 40% of knowledge worker tasks are fully explicit (formalized, described, measurable).
60 to 70% of work is implicit, what ergonomists call "actual work" versus "prescribed work." These implicit tasks include: managing unspoken relational dynamics, understanding the political context, adapting procedures to operational realities, and emotional regulation. This "glue" is what makes us human. It's what creates belonging, distinction, branding, experience, and culture.
So yes, AI is amazing for producing reports in a predetermined format, analyzing data tables, and performing trading. But the task doesn't make the job. If that's the case, you need to expand your job responsibilities. On the other hand, clients can quickly access generic market information (and information heavily biased by the overrepresentation of HR tech giants, without any local or regional nuance). We're already seeing this. Clients asking questions about HRIS choices are offered ultra-comprehensive reports that are heavily skewed by the question or the training data.
Any agentic AI project that relies solely on existing job descriptions often automates the wrong scope. This is precisely why an analysis of actual, not prescribed, work must precede any decision regarding an automation roadmap with agentic AI and an HRIS roadmap.
When you are sold your Digital Twins or your AI Employee, we are talking about an assembly of 3 to 4 short and distinct processes, poorly integrated but formidably efficient in several languages, 365 days a year, 24/7.
As McKinsey aptly puts it, we are becoming AI Agent Orchestrators . They take it a step further by adding the concept of " Human in the Loop " to that of " Human Above the Loop . " We must validate AI, but also frame it within an increasingly complex architecture that is never truly autonomous, regardless of what the AI gurus trying to sell you their services might tell you.
I had already written about our role as AI trainers several years ago. Here we are.
In the spirit of Yann LeCun, as paraphrased by Bo Wang on Twitter /X, February 17, 2026, "If language were enough to understand the world, we could learn medicine just by reading books. But that's not how it works. You have to do your residency. You have to have seen thousands of normal cases before recognizing the abnormal."
One of the most prominent researchers in the world of Deep Learning also raises something striking: all the public text available on the internet represents approximately 10¹⁴ bytes.
Therefore, it's important to understand that a 4-year-old processes roughly the same amount of information through sight alone. The world is simply... bigger than the text.
Most human knowledge, and almost all animal knowledge, comes from sensory experience of the physical world. Language is the icing on the cake. You need the cake first.
Actions to be taken by HR departments in the age of AI:
1- Understanding generative AI and agentic AI.
This is fundamental. AI isn't just for IT professionals. It presents economic, ethical, and societal challenges that impact HR. It's essential to understand your needs and the processes that can be automated (in terms of volume). It's also crucial to understand the gray areas that will never be automated or will only be so at a prohibitive cost.
As indicated in our 1st AI in HR Barometer in Canada , we are witnessing the adoption of AI in HR but we are swimming in " AI Confusion " .
AI is widely adopted by HR professionals as individuals, but its implementation by organizations is still limited or insufficient. Moreover, the role of Shadow AI, which has completely overwhelmed businesses and HR departments, is frequently highlighted.
2- Set up a monitoring unit directly linked to the management team.
This monitoring unit must monitor The new developments and some forward-looking analysis of the business impacts of operations and training needs. No one can live in a bubble, immune to the greatest technological acceleration in history, especially not executives.
Some of your field employees and consultants are testing these technologies for $20 to $70 a month before you even hear about them.
3- Equip, Test, Understand the limits. Calm down and think as a team.
Never before has the complexity of IT architectures, cybersecurity, and robotics demanded so much ingenuity and expertise. Jobs are being created.
Can we imagine a company without human labor?
4- Improve your ability to produce quality and secure HR data as well as up-to-date explicit knowledge (policies, procedures, in-house training).
The nightmare of " data chaos " will haunt you. No AI agents without quality data and knowledge bases! Your outdated and difficult-to-interface HRIS systems are broken bridges on the road to AI.
The real challenges facing HR departments with AI:
1-Build an intelligence infrastructure.
While we are only just beginning to transform organizations with digital technology, generative AI and agentic AI are raising the standards of intellectual productivity. The cost per report page, per customer response, and per line of code is approaching zero.
Successful companies in 2026 are skillful combinations of intangible and tangible capital: skilled people, tools and agents, processes, intellectual property, brand and reputation.
2- Quickly develop expertise in AI.
" The struggle is the skill ." The skills of technology and data specialists, as well as those of end users , are affected. The former are in high demand to gain speed. The latter's skills reveal a huge training gap.
The cost of AI licenses remains relatively high compared to the HR technology ecosystem. Equipping all white-collar workers with OpenAI ChatGPT, Microsoft Copilot, or Claude Pro costs $25 to $30 per employee per month. That's a significant expense when your HRIS, with its multiple corporate compliance uses, costs roughly the same in licenses. However, the return on investment for generative AI for your employees is certainly very positive.
3- Rethink the processes and the experience (really), don't just add a layer of AI.
The work of organizational architect and the hybridization of Human-Autonomous or semi-autonomous AI Agents is at the heart of the HR mandate by 2030.
Whether you like it or not.
To continue this discussion, download our survey.
"AI: Work Reinvented in Accelerated Mode" - 1st AI Barometer in HR in Canada:
Download page (French): https://bit.ly/baromètre-ia-rh-2026
Download page EN: https://bit.ly/1er-barometre-IA-en-RH-Canada
And what do you think?
How is your HR department monitoring and supporting this revolution? What initiatives have you already implemented?
Your comments, reactions and questions are welcome below.
Share this article with your colleagues and HR partners; transformation is a collective endeavor.
Reference :
I deliberately used the title of Yves Clot's book, "Work without man?" (La Découverte, 1995), in my article's opening. He is an emeritus professor of work psychology at CNAM and a researcher at the Centre for Research on Work and Development (CRTD).
I am referring here to one of my professional peers. The world of work is changing, and occupational psychologists are on the front line (cognition, training, attention, collaboration, decision-making).
About the author of this article:
This post is written by Jean-Baptiste Audrerie, co-founder of NexaRH, an independent consulting firm specializing in HR TECH strategy and digital transformation of human resources, active in Quebec, France and the United States.
Copyright of the 2026 Canadian HR AI Barometer:
© 2026 NexaRH inc. 1st Barometer of AI in HR in Canada “AI: Work reinvented in accelerated mode” .
All rights reserved.
Any reproduction, even partial, is subject to prior written authorization.
For reproduction or quotation requests: jb.audrerie@nexarh.com
Comments