Have your agent call my agent

Imagine every candidate had an AI agent of their own, not a careers chatbot they upload a resume to when they need a job, but something that has come to know them over time. It holds their work history, qualifications and learning records, along with any project work, documents and portfolio evidence they choose to connect, so it understands the skills they have, how they acquired them and what they could learn next. It remembers what they enjoy, what frustrates them, the money and working arrangements they need, and why they turned particular opportunities down. It is an informed representative of the person behind the resume.

Now imagine the employer has one too. It is built on business plans, product roadmaps, workforce data and what hiring managers and teams have learned from previous hires, not on a job description or a search of the ATS. It carries a live picture of what the work involves and what success looks like, and it can anticipate gaps before a vacancy becomes urgent.

When those two agents meet, they can do the whole front end of hiring between them. The employer’s agent describes a specific piece of work the team needs and asks what this person has done that shows they could contribute. The candidate’s agent answers from the evidence it holds. Together they explore the role, the working arrangements and the timing, and identify what still needs a human conversation. The candidate sees only the companies that are a great fit, the employer gets qualified, engaged people at the point it needs them, and the recruiter stops managing a funnel. Humans stay in control of the decisions, and the result is better conversations, more engagement and better hires.

This was the vision I set out in my presentation at RecFest USA in Nashville last month, and the evidence that it is coming is mounting. Consumer behaviour is where the trends show up first, and that is where the agents have already arrived. On 8 September Meta launched Muse, a personal agent that shops, books and manages email on its user’s behalf. It racked up millions of downloads across the US and Canada in its first three weeks. Companies have started to respond. John Lewis, the British department store, has launched Gift List, a YouTube chat show with content built specifically to help LLMs find and understand what it sells. The customer has changed how they choose, so the retailer has changed how it competes to be chosen, which now means marketing to the customer’s AI. We spend a lot of time asking how our AI should assess candidates. How much time are we spending asking how a candidate’s AI might assess us?

Recruiting is not far behind, and as usual candidates are setting the pace. They have tools that tailor and send applications at volume, and the next generation of those tools is starting to represent the person rather than optimise a single application. On the employer side, AI screening, AI-led interviews and talent intelligence are spreading, and vendors are building agents to represent the employer. Some of the early results show what becomes possible when the technology is used to understand people instead of processing them, with a better candidate experience and access to exceptional talent the old process would have missed.

Employers have a long record of reacting to this kind of change rather than leading it. When job boards arrived in 2000, candidates adopted them much faster than employers, who found themselves swamped with applications, very often from outside the local areas they had always recruited from. In a lot of cases the ATS was brought in to cope with that volume rather than as a strategic decision to improve recruiting efficiency. Almost every AI case study I see today follows the same pattern. Candidates started using AI to apply at a scale nobody had planned for, and employers bought technology to manage the volume and find the signal in applications that all look and sound the same. Each time, candidates have got there first and employers have been left to catch up.

In Nashville I asked where an employer’s AI ambition stops. Managing applications is the first level, and nearly all of the current activity sits there. Meeting the talent needs the business has now is the second. Shaping the capability the business will need next is the third, and it is the only one that changes when the conversation starts and what TA can influence. All three matter, but the employer agent I described can only be built at the third.

Left to the usual pattern, agent-to-agent hiring will arrive from the candidate side and employers will react to it, and there is plenty that could go wrong. Badly thought through implementations and poor technology would erode trust between candidates and employers further than it already has been, strip out the nuance that makes the whole idea worthwhile, and make bad recruiting processes worse at greater speed.

This is a question of readiness before it is a question of technology. Being ready means understanding candidate behaviour now, asking what you would build if you started from the hiring outcome instead of the application process you have, and settling who owns the safeguards in advance. Agent-to-agent hiring is going to happen whatever we do. Whether it makes recruiting better for everyone in it is up to us.