ATS: AI Agents Take Control
- Jean-Baptiste Audrerie
- Jun 5
- 9 min read
At the Talent Lab organized by Les Talents Narratifs and Laurent Brouat , Jean-Baptiste Audrerie presented his analysis of the market. This article summarizes and expands upon the points he presented during his conference on May 21, 2026, in Paris.
In addition to ATS, there are AI Agents.
The arrival of AI agents is revolutionizing work, organizations, and technologies. I was one of the first to discuss this in the HR community as early as January 2025 (HR Technologies Conference Paris), anticipating the impact on skills and employment.
What you used to do, they do : trigger a search as soon as a position becomes available, conduct interviews, pre-screen, move candidates through the pipeline, and even get offers signed.
The talent acquisition technology market has re-positioned itself around this concept of the agent in less than 12 months. This technological emergence can be observed in all other technological fields (finance, sales, customer relations, legal, engineering, etc.).
But above all, in addition to ATS technologies, there are individual commercial AI solutions , which provide individual efficiency, but are still rarely maximized for organizational productivity.
90% of HR professionals say they use a ChatGPT or Microsoft Copilot-type assistant daily. Most often, this use is covert ( Shadow AI ).
Between March and June 2026, more than half of the most tech-savvy professionals tested Claude Cowork, its Skills, and Plugins. They discovered the capabilities of "Computer Use." AI agents, acting on your behalf, connect to your applications via an MCP Server or API connectors.
Among the most advanced, users of Claude or Open AI ChatGPT have recently tested "vibe coding" (on-the-fly coding) in their living rooms.
1- It is important to distinguish between the conversational agent (chat mode) and the AI agent (cowork mode).
We need to distinguish between generative AI (conversational assistants and voice AI) and agentic AI (AI agents).
The agent takes action. They can access your browser, perform searches, schedule appointments in Outlook, and generate documents. Agents have learned to work and act in a digital environment by copying your actions.
2- If recruiters are not yet "agentized", candidates already are.
Your candidates are already using agents: Perplexity offers direct application automation. Dozens of plugins (freemium or premium) now exist for job tracking , keyword stuffing , autofilling , automatically generating resumes, or filling out ATS forms on behalf of candidates.
A tool like Simplify already knows 32 different ATS and can apply to hundreds of jobs per day on your behalf while you sleep.
In North America, this has led to a 50% increase in application volume in one year. 75% of applicants use at least one assistant or agent. This generates "perfect matches" on paper, but also " AI resume slop " (an avalanche of useless documents), fake applicants, and cheating during video interviews.
Some tools, such as the Canadian application Hyrr.ai , even train candidates with avatars to give excellent answers to recruiters, who are always more sensitive to verbal performance and presence than to contextualized competence.
3- The rapid commodification of the recruiter profession
On the recruiter side, we're moving from generative AI (writing job postings) to agentic AI (taking action). Some clients use agents integrated with Workday or SAP SF: as soon as a departure is announced, the system automatically launches a search within the internal talent pool and prepares a list of external candidates or a job posting if the replacement for the position is pre-approved.
The Applicant Tracking System (ATS) market is evolving. It's no longer about a single, monolithic "building block," but rather a multitude of agent-based capabilities. The recruiter thus becomes an orchestrator of agents. Moreover, for mass recruitment, automation can be almost complete.
4- The massive agencyization of the HR TECH market
SAP SuccessFactors acquired SmartRecruiters to gain agility with a suite of agents and add modern recruiting capabilities.
Workday acquired Paradox (a leader in mass recruitment by agents) to recruit with agents, without CVs and via mobile messaging.
Solutions like Deel or Remote now integrate sourcing and onboarding for companies that want to expand rapidly internationally with an EOR solution.
Not all agents are AI agents. Certainly, AI agents offer a wide spectrum, ranging from true agentic AI (memory effect, generative AI injection, semi-autonomous actions) and good old programmatic automation mechanisms (linear logic: if this, then that).
5- These three technological levels fit together. They are not mutually exclusive.
Today, you have three choices that coexist:
The conventional ATS (SaaS): This represents the foundation. It's the source of truth for data and security. Conventional ATS, especially for larger, multi-entity companies with strong compliance requirements, need onboarding processes, recruitment budget management, reporting, and security that demand established and robust solutions. They are relied upon to incorporate best practices and keep pace with regulations (GDPR, AI Act, etc.) and the evolving solutions within the HR TECH and TA ecosystem (APIs and MCP servers).
The AI Conversational Assistant: For engagement and capturing candidates quickly on the website (very effective for frontline employees), these conversational assistants facilitate the capture and recapture of prospective candidates or candidates from the talent pool.
AI agents: They enhance or complement existing solutions, or sometimes replace them for specific tasks. Some best-of-breed solutions are built around highly specialized use cases. Examples include note-taking, appointment scheduling, pre-selection interviews, skills assessment, and matching.
6- The ATS is the cornerstone of your "TA Stack".
Three strategies are available to you:
Buy (rent off-the-shelf) turnkey SaaS solutions. You rent capacity and pre-configured agents. You delegate management and maintenance to focus on usage.
Bridge (connect your systems, notably via Anthropic's MCP protocol).
Build (prototype yourself via vibe coding).
For those who are terrified by the idea that the job of recruiter could be totally "agentized", I present here the view of the technological range that makes up the TA Stack (Technological Environment in Talent Acquisition) that a large company or SME can theoretically have with growth and volume of recruitment.
It quickly becomes apparent that the range is vast, very vast. Replacing this entire application portfolio with AI agents like Claude Cowork or "Vibe Coder" is complex, risky, and expensive, unless it's your core business.
7- New practices, largely supported by AI agents
1- Towards mass recruitment becoming a commodity, largely agentized.
The rise of solutions like Paradox, Fountain, and others marks a turning point for the mass recruitment of frontline workers . This market is the ideal playground for AI agents, capable of automating the entire cycle, from onboarding to signing. By focusing on instant mobile access rather than traditional application forms and CVs, these tools reduce processing times from three weeks to four hours, radically optimizing productivity. The business model and the user experience are rapidly transforming.
2- Towards skills-based hiring
Degrees are becoming less important. In 2025, according to a LightCast survey, for recruitments with a surge in AI skills, 94% of hired candidates did not have the initially required degree but possessed emerging skills (such as vibe coding). Agents can now infer skills not mentioned in a CV by analyzing the candidate's career path.
3- Towards information production: the lever for differentiation and growth for recruiters and agencies of tomorrow.
Faced with the emergence of AI agents, we need to "data-ize" the process. If you can generate new and useful data and insights for candidates and companies, you move beyond the risky zone of purely transactional processes potentially monopolized by AI agents.
Moving towards informational data means taking a leading position: intimately understanding and documenting the "market-candidates-clients" context; transforming historical data, context, and results into indicators; providing targeted and personalized feedback; and strategically managing and guiding the process. Thus, creating data means generating value that AI cannot simply replicate. It means escaping the trap of the reactive order taker and the execution that risks becoming agent-like.
Whether internal or external, carrying out transactions can be codified and therefore monetized, at a certain cost. If your recruitment process is a series of transactions, your value tends towards zero.
8- Strategies and future of the recruitment profession: replaced (not really), recomposed (absolutely).
"An AI agent is not enough to replace a job, unless your job is reduced to a single task." - Jean-Baptiste Audrerie
AI agents take action on one or more tasks , but rarely, at least for now, on all the tasks of a corporate recruiter or agency recruiter job.
The recruiter position, according to the ESCO, O*net and LightCast job description database, represents between 135 and 175 tasks.
Creating an end-to-end AI agent chain is almost impossible, given the sheer number of parameters and the complexity and iterative nature of intangible and relational components.
One final point: even if theoretically possible, the current cost of tokenized agents and the required memory would make the project prohibitively expensive and futile, unless it captures enormous volumes of unique recruitment opportunities. Neither Google nor OpenAI, which announced their intention to automate the matching of employers and candidates, have achieved this.
9- Pilot, Co-pilot, Autopilot
My conviction and experience tell me that jobs are less being replaced than restructured. This restructuring of jobs with the arrival of agent-based AI is happening through the tasks themselves.
Each job is broken down into tasks which we classify as: Pilot, Co-pilot, Autopilot (PAC model).
Pilot: For relationship building, advice, internal policy and high-level expertise (approximately 50 to 60 essential human tasks).
Copilot: To assist you in writing and analysis.
Autopilot: For administrative and repetitive tasks (10 to 20 agents).
At Nexa RH, this is how we approach job descriptions and operationalize AI transformation, task by task. This makes things immediately more realistic and tangible: What tasks? What unique human added value? What AI? What template and process? What volumes? What costs?
The recruitment profession also needs to become more premium. If you remain focused on simple transactions, agents will replace you because the cost of technology is approaching zero.
Recruiters are not going to disappear, but they will need to develop their skills in the "pilot" aspect to distinguish themselves in what AI cannot handle.
10- Amid the "agentic AI" hype, remember this:
The job of recruiter is a highly iterative , context-dependent , and sometimes highly political and relational job (meaning: difficult to codify), especially for single-position processes.
Value creation is achieved by delegating certain well-targeted tasks to AI agents, combined with the enhancement of tasks with high human value (you define these).
In AI, reinvention is continuous (every 3 to 6 months). Your capital lies in proprietary data, formalized processes, and a high-level Human-Agent alliance.
Taking control of your employment is work.
Orchestrating swarms of agents requires codifiable transaction volumes .
Agentic AI is expensive (security, tokens, memory, governance, controls, continuous improvement, changing or upgrading models).
Good Agentisation.
Good Data-ification.
Good Premiumization.
To learn more:
Read our article: HR Tech Market 2026: Consolidation and Agentization.
Read our article: The 32 Mergers and Acquisitions of the HR TECH market 2025 .
Read our article: Towards an agentic organization
Read our article: Examples of AI Agents
Glossary of technical terms and concepts
AI represents a whole new universe of concepts and notions. The language is evolving rapidly.
Agentization (Agentizing) : The transformation of a job, process, or organization through the delegation of tasks to AI agents. "Agentizing" one's work means analyzing tasks one by one to entrust the most codifiable ones to agents and elevate tasks with high human value.
Agentic AI: AI capable of acting semi-autonomously: chaining actions, connecting to applications and executing tasks on your behalf, beyond simply generating content.
AI Agent: A program that operates in a digital environment (browser, software) by copying human actions: searches, appointment scheduling, document production.
"Computer Use": The ability of an AI agent to directly control a computer (keyboard, mouse, screen) as a human would, to use applications not intended for automation.
MCP (Model Context Protocol) Server : Anthropic's open protocol allowing an AI agent to connect in a standardized way to external applications and data sources.
"Tokens" : Units of text (words or fragments of words) that an AI model processes. Token consumption measures and charges for the use of AI agents: the longer or more repetitive a task is, the more tokens it costs.
"Vibe coding" : Coding on the fly: developing a prototype or application by describing in natural language what you want, with AI generating the code.
EOR (Employer of Record) : A legal third-party employer that hires employees on behalf of a company, particularly for international expansion without a local entity.
"Frontline Workers" : Frontline employees (field, service, production), often recruited en masse and continuously.
"AI resume slop" : An avalanche of CVs and applications generated en masse by AI, often of low value, which saturates recruitment processes.
Data-ification (Data-izer) : A strategy that, in the face of AI agents, consists of creating new and useful data and insights for candidates and companies, in order to move beyond purely transactional processes and create value that AI cannot simply replicate.
Premiumization (Premium-iser) : Upgrading the recruitment profession towards tasks with high human value (relationship building, advice, strategy), as opposed to simple transactions that AI agents can automate at almost no cost.
PAC model (Pilot / Co-pilot / Autopilot) : NexaRH framework for the distribution of tasks within a job: Pilot (essential human tasks), Co-pilot (AI assistance) and Autopilot (tasks automated by agents).
Looking forward to reading your comments or questions.
If you would like to organize a conference for your AT teams on this topic, please contact me via the website form .
Comments