Click here to get this post in PDF
In 2026, hiring decisions are increasingly being led by a simple question: do we need a person at all?
Work involving relationships, context, judgement and accountability invariably relies on people. Repetitive tasks and defined workflows are different, and AI agents increasingly give businesses a choice about how that work gets done.
The question isn’t one of replacing the human workforce or, alternatively, of maximising agent deployment. It’s about understanding where each belongs. People for people. Agents for process.
The Team You Don’t See on the Org Chart
A conventional org chart still tells us who works for a business, but it’s becoming less reliable at telling us how the work gets done.
Behind the sales manager and customer service team might now sit agents qualifying leads, resolving routine enquiries, updating records or moving information between systems. They don’t appear in the headcount, but they contribute in a meaningful way to operating capacity.
That changes the hiring equation. More work no longer automatically means more employees. If an agent can absorb the additional workload, the existing team can focus its energies elsewhere. A salesperson still builds the relationship and owns the outcome, but an agent might research hundreds of prospects, qualify opportunities and keep the CRM current around them. The most effective teams aren’t necessarily human or automated. They’re both.
It’s Not Just About Cost
While AI agents can provide cost savings, framing them as ‘cheap labour’ undermines their true value. Agents can process work as it arrives, perform repeatable tasks consistently and absorb increases in volume without the recruitment and onboarding required to expand a human team: ten requests can become a thousand without any need to increase headcount. For process-heavy work, that combination of speed, consistency and scale matters easily as much as cost.
Agents are Joining the Enterprise
We are already seeing the impact of this shift in the way major enterprise platforms are being built. Salesforce’s Agentforce, for example, has been designed around agents that can reason across workflows and take actions within business systems, rather than an AI that simply delivers information for a person to act on.
The broader significance is that enterprise software is increasingly being designed around people and agents working alongside one another. As that becomes embedded within normal business infrastructure, the challenge itself shifts from whether or not agents can work at all, to understanding how businesses can put them to work properly.
If It’s a Hire, Treat It Like One
Nobody would recruit a human employee, offer them unrestricted access to company systems, point vaguely in the direction of a department in which they might like to work and hope for the best. Yet, businesses stand to do exactly that if they choose to deploy agents without clearly defining their role, authority or accountability.
If an agent is going to perform work that might otherwise have required a hire, its implementation needs to begin in much the same place: with a clear understanding of the job. This begins with defining a role, rather than presenting a loose collection of ambitions: “handle order enquiries across all channels, escalating refunds above £500” gives the agent boundaries that can be understood and measured. A vague instruction, like “answer customer questions” leaves role scope problematically wide open.
Authority needs equally clear limits. What systems can it access? What can the agent actually decide? What requires higher permission? These are boundaries that need to exist in permissions and controls, and not exclusively in instructions provided to the agent.
Onboarding, for a human employee, involves learning how the organisation works. For an agent, much of that context can be drawn out from company data. But, if fragmented records, conflicting information and outdated documentation don’t disappear when AI is introduced, then that will be the material from which it works.
And every agent needs a manager. Someone to own its performance, review exceptions and intervene when something isn’t working. An agent with no named owner isn’t an autonomous colleague. It is an unmonitored system with production access.
Finally, you should decide how success will be measured before deployment. Resolution rates, accuracy, escalation rates, cost per outcome and the human hours released are all more useful than simply counting how many tasks the agent completed. If an agent is going to join the team, it should be managed like it’s really part of one.
What Are You Actually Hiring the Person For?
Once process can be separated from people, the case for a traditional hire becomes stronger rather than weaker. The strongest case for hiring a person exists where the value of the role comes from something that’s difficult to turn into process. One useful test for this is CROJ: Context, Relationship, Ownership and Judgement.
Context is the knowledge that never quite makes it into the CRM. The why of a particular decision; the understanding around which client needs careful handling, and what happened the last time you tried something similar. Relationship is the value created because somebody knows the customer, and the customer knows them. Ownership is the difference between completing a task and being answerable for the outcome. And Judgement is what happens when there isn’t enough information, when the rules don’t quite apply but somebody still has to make a call.
In most businesses, it is in these areas that people create the greatest value in the first place. Understanding that changes the hiring calculation again: the more heavily work depends on context, relationships, ownership and judgement, the stronger the case for a person. The more it depends on repeatable process, measurable outcomes and scalable execution, the stronger the case for an agent.
The opportunity isn’t to replace people with agents at all. It’s to stop hiring people for the tasks that no longer need them.
Satish Thiagarajan is the founder and CEO of Brysa
You may also like: Space raises $2.4M led by a16z Speedrun to build the AI-native filesystem for humans and agents
Image source: Satish Thiagarajan
