The Adecco Group is extending Salesforce’s Agentforce Coworker across more than 40 countries, taking recruitment automation that began with narrower candidate workflows into the daily work of around 27,000 employees. The deployment follows pilots in the UK and France and builds on recruitment agents already operating in ten countries that account for roughly half of group revenue. Its scale will test whether agentic AI can move beyond isolated tasks while still functioning across different labour markets, data systems, languages, and established processes.

Coworker is embedded in Salesforce and is intended to give employees a common interface into information and tools that previously sat across several systems. Sales staff will be able to identify prospects and prepare account information, while recruiters can invoke specialised agents for activities such as pre-screening and onboarding. Adecco says the system will also draw on experience from millions of interactions between candidates and its existing automated services.

Those earlier deployments include agents used for candidate screening, shortlisting, scheduling, and documentation, giving the group operational data before the larger employee rollout begins. Adecco has also reported that automated pre-screening handles substantial candidate activity outside normal office hours in the UK and France. The new programme is therefore an expansion of an existing operating model rather than an AI pilot starting with a blank workflow.

The scope changes as a common interface begins coordinating several agents rather than performing one bounded task. Staff no longer need to understand which application holds every piece of information, but the system underneath still has to resolve identities, permissions, data quality, and conflicting records across a multinational estate. A simpler interface can disguise fragmentation; it does not make the underlying integration problem disappear.

Automation becomes a governance problem as it acts

The risk profile changes when an AI system moves from retrieving information to triggering workflow. Preparing a sales briefing is relatively contained, whereas selecting candidates for screening, starting onboarding, or updating operational records can affect people and downstream systems. A deployment across more than 40 countries therefore needs controls over which actions an agent may take, what evidence employees can inspect, and where local processes override a global template.

Recruitment adds another layer because automated recommendations can influence employment outcomes. Adecco says human judgement remains part of its operating model, although the useful measure will be how recruiters actually use generated rankings, screening results, and suggested actions. As automation handles more early-stage work, organisations need to know whether human review remains substantive rather than becoming routine confirmation of the system’s output.

Scale at least gives Adecco enough volume to measure those questions. Error rates, recruiter overrides, candidate completion, response times, service quality, and the distribution of problems between markets can all be examined once the same technology reaches meaningful operating volumes. Those measures provide a stronger basis for judging performance than adoption figures alone.

Multinational deployment also exposes the limits of standardisation. Candidate expectations, job classifications, data availability, local recruitment practice, and regulatory requirements vary considerably, while agentic systems are often sold partly on their ability to work across several tools without forcing staff through rigid processes. Adecco will consequently be testing whether a common layer can absorb local variation or simply create another centrally imposed system regional teams have to work around.

The integration layer becomes strategic

An enterprise agreement can remove some commercial barriers to experimentation by making AI capabilities broadly available, but it also changes the governance task. Once employees can invoke agents from a common interface, control moves away from deciding whether a department may use AI and towards defining which data, systems, and actions each agent can access. Permission design becomes part of application architecture rather than an administrative afterthought.

The same dynamic is appearing across large enterprise deployments as organisations accumulate multiple specialist agents. Individual tools may perform useful tasks, yet employees need a coherent way to discover and coordinate them without creating another fragmented software estate. A common interface can solve part of that problem only if the organisation has done the less visible work of cleaning identities, permissions, integrations, and data flows underneath.

Adecco is already far enough into deployment that the programme can be evaluated as an operating change rather than an innovation exercise. The 40-country expansion puts agentic systems into the work of tens of thousands of employees and connects them with candidate and client processes that generate measurable outcomes. The harder evidence will come from whether the technology reduces fragmented work without introducing new errors, opaque decisions, or extra layers of supervision.