For most of the software era, the distinction between technology and labour was clear. Companies hired people to perform work and bought software to help them perform it. A CRM did not occupy a role in the organisation. It stored information and provided functions that employees learned to use. If the company wanted more sales capacity, it hired salespeople. If it wanted a better sales system, it bought software.
Agents break that separation.
An agent can be given an outcome rather than a single instruction, work across several applications, retain context, communicate with people and other agents, and carry a sequence of work forward without requiring a human to initiate every step. The word “agent” is still being applied rather generously across the industry, but the direction is clear: software is moving from tool to participant.
That change will fundamentally alter the commercial model around software. Companies will still buy models, platforms and applications, but access to an agent is becoming the easy part. The difficult part is selecting the right agent, integrating it into a real team, teaching it how the organisation works, establishing boundaries and responsibilities, helping human colleagues work with it, monitoring its behaviour, improving it over time and taking responsibility when it underperforms.
This creates a new business category between software, staffing, consulting and organisational design: the Agent Staffing Agency, or ASA.
The phrase is deliberately provocative. Agents are not employees, and treating them as human would obscure important differences. But staffing provides a better analogy than software because staffing agencies do something software vendors generally do not: they take responsibility for placing a capability into an organisation and helping ensure that it works there.
The shift is simple:
From SaaS to ASA.
Software as a Service gave companies access to capabilities without requiring them to build the underlying technology. Agent Staffing Agencies will give companies access to operational agent capability without requiring every organisation to become expert in recruiting, onboarding, integrating, governing and continuously developing digital team members.
That distinction is much larger than it first appears.
An agent can be right and still be a terrible team member
Traditional software support rests on a relatively simple assumption. The software is either working as specified or it is not. A function executes correctly, produces an error, behaves unexpectedly or fails altogether. Support diagnoses the problem against an expected technical state.
Agents introduce a different class of failure.
An agent can be technically functioning and still be a terrible colleague.
It can give correct answers while communicating them in an unhelpful way. It can escalate too often and become a burden on the team, or fail to escalate when human judgement is required. It can understand its own task perfectly while ignoring how its actions affect neighbouring functions. It can be overly literal, intrusive, verbose, timid, excessively confident or obsessed with completing its objective at the expense of the broader purpose of the work.
None of these necessarily mean the software is broken.
The agent may be doing precisely what it was configured to do.
The problem is that the agent does not fit the organisation.
This distinction is fundamental. Software support asks whether the system works. Agent support asks whether the system works well with others.
That is a different discipline.
A technically excellent agent that irritates every employee, floods channels with notifications, ignores social context and creates constant cleanup work is a failed deployment. So is an agent that generates excellent individual outputs while undermining a team’s decision process. An agent that is right most of the time but destroys trust around itself can be less valuable than a less capable agent that understands when to speak, when to wait, when to ask and when to return the matter to a human.
Once agents participate in organisational life, performance becomes relational as well as technical.
That makes onboarding central.
Onboarding becomes the heart of agent integration
Companies already understand that hiring a talented person does not automatically produce a successful employee. Even the best recruit has to learn the organisation: its purpose, products, customers, language, systems, informal norms, decision pathways, risk tolerances and peculiar ways of getting things done.
Agents need the same kind of integration.
A capable agent arriving inside an unfamiliar company does not know which information matters, whom it should consult, what authority it has, what constitutes an exception, how customers should be addressed or which organisational values should govern difficult trade-offs. It may understand accounting, marketing or procurement extremely well while knowing almost nothing about how this particular organisation practises accounting, marketing or procurement.
Onboarding therefore goes far beyond technical configuration.
An agent needs to understand why the organisation exists and what outcomes matter beyond immediate efficiency. It needs to learn the language and conventions of the team it joins. It needs explicit permissions, escalation paths and boundaries. It needs access to the right knowledge without indiscriminate access to everything. It needs to understand who has authority over which decisions and how changes elsewhere in the organisation affect its own role.
Most importantly, onboarding happens in both directions.
The humans also need to learn the agent.
They need to understand what it does well, where it is unreliable, how to give it useful context, when to challenge its recommendations and how its role interacts with their own responsibilities. Teams need working norms around how much autonomy the agent receives and how its activity becomes visible. Managers need ways to evaluate its contribution without pretending conventional software metrics are enough.
This is why agent integration is becoming a substantial organisational discipline in its own right. It is not installing a new application and running a training webinar.
It is integrating a new kind of participant into the operating system of the organisation.
HR for agents
The staffing analogy becomes even more useful here.
A contemporary staffing agency does not merely provide a list of candidates. It helps define roles, source talent, evaluate fit, handle placement and remain involved when a placement fails. The best agencies understand that capability and fit are different questions. Someone can be exceptionally skilled and completely wrong for a particular organisation.
Agent ecosystems create the same problem in a new form.
Companies need to determine which type of agent belongs in which role, what capabilities it requires, which underlying model or platform is appropriate, what permissions it should receive and how it fits alongside existing people and systems. Agents need to be evaluated not only for accuracy and speed, but for reliability, communication style, degree of autonomy, ability to explain uncertainty and compatibility with the organisation’s way of working.
This becomes a new form of HR for agents.
There will be agent role design, selection, onboarding, training, performance evaluation, development, reassignment, retirement, access management and behavioural adjustment. There will be conflicts between agents and humans, and increasingly between agents themselves. There will be questions about who an agent reports to, what it is authorised to do, how its performance is judged and what happens when its role changes.
Some of these functions are highly technical. Others are fundamentally organisational.
The Agent Staffing Agency sits directly across this boundary.
It can source an existing agent instead of building one from scratch. It can customise that agent for the client, integrate it with internal systems, supervise its onboarding and help the human team establish working practices around it. It can monitor performance, retrain or replace the agent when necessary, and remain accountable for the quality of the deployment.
The value is not access to intelligence.
The value is confidence that the intelligence has been responsibly integrated into the organisation.
Warranty changes when the product has behaviour
This also changes what support and warranty mean.
When companies buy conventional software, they expect technical support. If the application crashes, behaves incorrectly or fails to deliver a contracted feature, the vendor fixes it.
Agent support is different because the problem is often behavioural rather than technical.
Suppose the agent technically works but continually misunderstands the company’s priorities. Suppose it completes tasks successfully but communicates with customers in a way that damages the brand. Suppose employees stop trusting it because its advice is correct in isolation but ignores context everyone inside the team considers obvious. Suppose an agent performs beautifully for three months, then a change in the underlying model alters its behaviour.
These are not conventional bugs.
They are still service failures.
An Agent Staffing Agency can therefore provide something closer to a fitness-for-purpose warranty. It cannot guarantee that an agent will never make a mistake, any more than a staffing firm can guarantee that a person will never make one. What it can guarantee is continued responsibility for whether the agent remains suitable for the role.
That means monitoring behaviour, investigating failures, tuning instructions and memory, adjusting permissions, changing integrations, replacing underlying components, redefining the role or replacing the agent itself.
The agency stands behind the placement.
That is considerably more valuable than technical support.
The provider is effectively saying: we will not simply give you an agent; we will take responsibility for making that agent work inside your organisation.
That responsibility is the core commercial proposition.
From individual agents to managed teams
The model becomes more powerful when the unit of delivery shifts from individual agents to agent teams.
Most significant organisational work crosses several capabilities. A customer-service process can involve one agent retrieving relevant cases from a CRM, another searching technical documentation, another comparing possible solutions with previous outcomes, and another preparing an explanation appropriate for the customer. A human support engineer remains responsible for judgement and relationship, while a group of agents expands what that person can perceive and accomplish.
Marketing can involve a research agent, an audience-intelligence agent, a campaign agent, a content agent and an analytics agent. A strategic function can combine agents specialising in market signals, internal knowledge, scenario analysis and organisational memory.
The valuable product is not any one agent.
It is the quality of the human-agent team as a system.
This changes the design question. Companies stop asking only which AI tool they should buy and begin asking what configuration of humans, agents and applications should perform a particular function.
That is organisational design.
It also means someone must manage the relationships among agents. One agent’s output becomes another’s input. Different agents operate with different permissions, models and assumptions. Their roles overlap, contradict or leave gaps. A small change in one part of the system can create consequences elsewhere.
The organisation therefore needs more than agents doing work.
It needs agents that help agents work.
The rise of agent-support agents
This will become one of the most specialised parts of the field.
As companies operate larger populations of agents, they will need meta-level agents whose role is not to perform conventional business functions but to support the agent workforce itself.
An onboarding agent can introduce a newly deployed agent to organisational purpose, policies, systems, terminology, permissions and relevant collaborators. A monitoring agent can observe patterns of performance and surface unusual behaviour. A coordination agent can maintain awareness of which agents are responsible for which functions and route information between them. A governance agent can watch permissions, escalation rules and policy boundaries. A learning agent can identify recurring failures and propose improvements across the agent system.
These are not generic assistants.
They are the infrastructure that makes other agents useful.
The analogy to HR becomes stronger again. Human organisations contain people whose work largely consists of helping other people join, coordinate, develop and perform well. Agentic organisations will develop an equivalent layer.
But these agent-support functions cannot simply be automated and abandoned to themselves. They need to work closely with human consultants, managers, technical specialists and organisational designers.
An onboarding agent can hold organisational context, but humans still decide what the organisation’s values mean in practice. A governance agent can detect an unusual permission pattern, but a human may need to judge whether it is justified. A coordination agent can identify overlapping responsibilities, but managers still need to decide how authority should be distributed.
The emerging profession is therefore much broader than AI engineering.
It is the design and stewardship of human-agent organisations.
Integration becomes the scarce skill
The AI industry is currently obsessed with capability. Which model is smartest? Which agent can use the most tools? Which benchmark has improved?
Those questions matter, but they will become progressively less decisive as powerful models and pre-built agents proliferate.
Integration becomes the scarce skill.
A company can acquire a capable agent easily and still have no idea what to do with it.
Enterprise software has already taught this lesson. Organisations routinely buy sophisticated systems and then discover that the technology was the least difficult part of the transformation. Processes are unclear. Data is fragmented. Departments disagree. Nobody owns the workflow. Employees resist the change. The software is operational while the organisation remains stubbornly itself.
Agents raise the stakes because they do not simply sit inside workflows.
They act inside them.
Companies therefore need people who understand technology and organisational change simultaneously. They need practitioners capable of examining a human workflow, recognising where agents belong, redesigning roles and information flows, configuring the right agents, building governance around them and helping human teams establish new working relationships.
Most organisations will struggle to build this capability internally.
The field moves too quickly and the skill set crosses too many existing functions. It combines AI architecture, process design, organisational development, change management, HR, knowledge management, governance and coaching.
Large companies will build substantial internal agent-operations departments.
Most companies will not.
They will rely on external specialists.
That is the opening for the Agent Staffing Agency.
The managed-service model
The commercial architecture follows directly from the operational problem.
An engagement begins with diagnosis. The provider examines the organisation’s goals, workflows, systems and current pain points, then identifies where agents can genuinely improve the work. It moves into role design, agent selection or development, configuration, integrations, permissions and security.
But the most valuable phase begins after the agent technically works.
The agency onboards it.
It introduces the agent to the organisation and the organisation to the agent. It works with human teams to establish interaction patterns, escalation rules and expectations. It observes what happens when the neat process diagram collides with actual organisational life.
Then it stays.
As the organisation changes, the agents change with it. New employees arrive. Products evolve. Policies shift. Data sources move. Agents gain new capabilities. Underlying models improve. An agent that worked well six months earlier may no longer be the right choice.
The relationship therefore becomes a managed service rather than a completed software implementation.
Clients pay a recurring fee covering agent management, monitoring, upgrades, retraining, replacement, new integrations, behavioural tuning, governance and human support. The agency guarantees continuity of capability rather than permanence of a particular technical implementation.
That is the key distinction.
The client does not care whether version 3.7 of Agent X remains in place forever.
It cares that the research, support or coordination capability continues to work.
A staffing agency does not promise that one particular person will occupy a role forever. It promises access to appropriate talent and continuity of service.
ASA follows the same principle.
From workforce reduction to organisational expansion
The least imaginative version of agent staffing is labour substitution. Identify jobs, insert agents and reduce headcount. Some organisations will pursue exactly that model.
But capability expansion is the larger opportunity.
A medium-sized company may never have been able to afford dedicated multilingual research, sales intelligence, knowledge management and customer support teams. Managed agents make those capabilities accessible without requiring the company to build full departments around them.
A small consulting firm can use agent teams to handle research, knowledge retrieval, proposal preparation and project coordination, allowing human consultants to spend more time on judgement, relationships and complex client work.
A nonprofit can gain research, communications, fundraising analysis and organisational memory capabilities previously available mainly to much larger institutions.
The question changes from “Which people can we remove?” to “What can this organisation now become capable of doing?”
That is a much richer proposition.
It also places the Agent Staffing Agency in a different category from an automation consultancy. The objective is not to minimise the number of humans. It is to design a stronger combination of human and machine capability.
The deeper asset is learning how agents join organisations
As the model scales, the most valuable intellectual property will not be the agents themselves.
Imagine a provider integrating agent teams into fifty organisations. Over time, it learns where implementations fail, which onboarding patterns work, how much autonomy different teams tolerate, which permissions create unexpected problems, where employees lose trust, which agent combinations cooperate well and which behavioural patterns repeatedly create friction.
That knowledge becomes extraordinarily valuable.
A conventional staffing firm develops expertise in matching humans to organisational roles. An Agent Staffing Agency develops expertise in matching and integrating agents into organisational systems.
Some of that learning will eventually become embodied in the support agents themselves. An onboarding agent becomes better at understanding new organisations. A coordination agent recognises common patterns of role confusion. A governance agent learns which configurations deserve human attention.
Every deployment can improve the next one, provided privacy, ownership and consent are carefully protected.
The underlying models will increasingly commoditise.
The accumulated capability to make agents productive, trusted participants in real organisations will not.
A new category of organisational infrastructure
The Agent Staffing Agency sits at the intersection of several familiar industries without fitting neatly into any of them.
It resembles SaaS because digital systems provide capabilities continuously. It resembles staffing because the customer acquires an operational role rather than merely a tool. It resembles consulting because successful deployment requires understanding and redesigning the organisation. It resembles managed services because the provider remains accountable after implementation. It resembles HR because agents require selection, onboarding, development and performance management.
But the combination creates something new.
ASA is not simply a more convenient way to acquire agents.
Its deeper function is to become the intermediary between rapidly evolving artificial intelligence and organisations that are not designed to absorb it unaided.
There will be no shortage of capable agents. Building them will become easier. Finding them will become easier. Model capability will continue to rise.
The scarce capability will be turning those agents into reliable participants in living organisations.
Someone has to understand what role is actually needed. Someone has to find or configure the right agent. Someone has to introduce it to the organisation. Someone has to help humans learn to work with it. Someone has to observe how the relationships evolve, improve the design and intervene when things go wrong.
And someone has to manage the agents whose job is to help all the other agents work well together.
That is not software support.
It is organisational infrastructure for a hybrid workforce.
The defining promise of the Agent Staffing Agency is therefore surprisingly familiar: we do not merely provide the capability; we stand behind the placement.
From SaaS to ASA, the decisive shift is from selling software to taking responsibility for integration.
As agents become more capable, that responsibility becomes more valuable than the agents themselves.
First published on LinkedIn, 15 September 2026.
← All posts