There is a version of this article that starts with fear.
It tells agents to be worried. It tells BPOs to move faster. It quotes a big consulting firm, points to some jobs that are already disappearing, and then wraps it all up with the usual "adapt or be left behind" message that everyone has already heard a hundred times.
This is not that version.
Not because the fear is wrong. Some of it is absolutely justified. If your work today is mostly reading from a script, answering the same question over and over again, or routing customers to the right queue, then yes, that work is being automated. Not in some far-off future. Right now. I think we need to be honest about that.
But I also think the industry has spent too much time talking about what AI will replace, and not nearly enough time talking about what we are supposed to build next. That is the part that interests me.
The question is no longer whether AI will change the agent role, it already has. The real question is what the agent role becomes after AI takes the easy work away, and whether we are actually preparing people for that version of the job. Right now, I am not convinced we are.
And I say that as someone who has spent a lot of time inside contact centers, BPOs, training rooms, calibration sessions, QA reviews, and production floors. I have watched agents do work that people outside the industry do not fully understand. I have watched them listen to emotional customers, search across broken systems, translate confusing policies into human language, calm people down, protect the company, and somehow document the whole thing before the next call lands in their headset.
That was never low-skill work, we just treated it that way too often, and now AI is exposing the mistake.
AI Is Not Coming for All Agent Work. It Is Coming for the Easy Work First.
AI is very good at work that is predictable, repetitive, contained and delivered with consistency that humans can’t replicate easily.
It can answer the same question at 3 a.m. the same way it answers it at 3 p.m. It does not get tired, irritated, bored, distracted, or emotionally worn out after a long shift. It can check an order status, reset a password, explain a basic policy, process a refund, summarize account history, or walk someone through a troubleshooting step.
We have had automation in contact centers for years, so none of this is completely new. The difference now is that AI can work with much more flexibility than the old rules-based tools. It can understand context better, adjust its response, interpret language, and handle more variation if the systems behind it are set up properly.
For operations leaders, this is not hard to understand. Every year, they are asked to reduce cost, improve productivity, protect service levels, and somehow make the customer experience better at the same time. AI is going to be part of that equation because the pressure to take cost out of the business is not going away.
That is the hard fact, as is this other hard fact: AI still struggles when the situation becomes messy.
Customers come in with hard challenges, they are not calling to make anyone’s lives easier, only their own. A customer calls about a billing issue, but the real problem is that they have already contacted the company three times and nobody has taken ownership. A policy says one thing, but the company clearly made a mistake and there is no flexibility in that policy, and an agent doesn’t want to be ‘marked down’ in QA so they just follow the process. A customer is angry, not because the issue is complex, but because the handoff before the human agent was terrible. Someone speaks quickly, uses local expressions, explains the problem badly, and gives information out of order. The system says the case is simple, but the agent can tell there is something else going on.
That is where the human agent still matters, and not because humans are better at everything, we are not.
Humans matter because some situations require judgment, emotional awareness, communication, and accountability. The system can suggest an answer. The agent still has to decide whether that answer actually fits the person in front of them.
That is the work we should be preparing agents to do.
The Easy Work Is Going Away. The Remaining Work Is Harder.
I think this is where a lot of AI strategy gets lazy. There is an assumption that if AI handles more customer interactions, the human agent role becomes simpler or smaller. I don’t see it that way and it’s definitely not always the case.
The volume may get smaller, yes. There may be fewer simple calls going to humans. But the work that remains will not be easier. It will be more complex, more emotional, more ambiguous, and probably more exhausting if we do not train people properly. Your typical BPO clients don’t typically factor in program complexity, they just expect superior delivery. So whether it’s 5K easy calls or 500 hard ones, there isn’t always the recognition that agents might need more help other than a lengthy knowledge document.
By the time a customer reaches a human agent, they may have already tried the website, the app, the chatbot, the help center, the automated email, or another department. They may have already repeated themselves. They may already be annoyed before the agent says hello.
So the agent is not starting from zero anymore. They are stepping into an interaction that is already in motion. That changes the job and where leading with empathy is more important.
The agent now has to read the context quickly, understand what the customer already tried, figure out where the system failed, recover the conversation, and solve the issue without making the customer feel like they are starting all over again. And they should have an idea of why this issue matters to the customer, and where their headspace might be in the moment.
That is not basic customer service, that is a higher-order skill set that many don’t possess, especially if this job is their first out of university.
And yet, in many places, we are still training agents like the main goal is to follow the process, say the line, hit the quality checklist, and avoid making a mistake. That model is not good enough for where the work is going.
We Have Overtrained Process and Undertrained Judgment.
This is one of the things that frustrates me most about the industry. We are very good at training agents on process. We can train product knowledge, systems, escalation paths, compliance steps, scripts, wrap codes, call flows, and knowledge base navigation. We can measure those things. We can test them. We can put them into a learning management system and track completion.
And to be clear, that training matters. I am not suggesting we throw it away. But process training is not the same as preparing someone for the real work of handling customers when the answer is not obvious.
I have seen agents who know the policy perfectly and still cannot handle the call well. I have seen agents pass training and then freeze on the floor because the customer did not behave like the roleplay customer. I have seen people with strong English on paper struggle when a live customer speaks quickly, interrupts them, uses idioms, or gets emotional.
That is not because those agents do not care, it is because we often train in controlled environments and then act surprised when performance changes in uncontrolled ones.
A training room is not the floor. A roleplay with unlimited thinking time is not the same as a live call with a frustrated customer, three systems open, a queue building, and a supervisor asking why your handle time is high. This is why future-proofing the agent workforce has to be about more than teaching people what to say. We have to teach them how to think under pressure.
Cognitive Load Is the Skill Gap We Barely Talk About.
If I could get more BPOs and contact centers to take one concept seriously, it would be cognitive load, because this is the actual job. An agent is listening to a customer explain a problem badly while also searching account history, waiting for a system to load, checking a policy, reading notes from a previous interaction, deciding whether the customer qualifies for an exception, watching the clock, managing tone, making notes, and trying not to miss the next thing the customer says.
That is not multitasking in the casual sense, this is the work they do every day that looks like they are on autopilot. I have spent enough time doing side-by-sides to know how impressive this can be. I have watched agents copy, paste, search, listen, apologize, explain, document, and make decisions almost simultaneously. And I have also watched what happens when the load gets too heavy.
The first things to disappear are usually the things we claim to care about most. Empathy drops, listening gets thinner, judgment gets weaker. The agent falls back on the script, or the tone becomes defensive. The customer feels managed instead of helped.
It is not always because the agent is bad. Sometimes it is because the agent is overloaded.
I once worked in an environment where agents had to use 17 different systems to solve different problems. Seventeen. We can talk all we want about empathy and customer obsession, but if the environment we put agents into is that complex, we should not be shocked when human performance starts to bend under the weight of it.
And now, as AI removes more of the easy interactions, the calls left for humans may carry even more complexity. That means cognitive load is not going away. It may get worse. So we need to stop pretending this is just a hiring issue.
It is a training issue, a tools issue, a leadership issue. It is a future-state workforce design issue.
Judgment is a skill and we need to train it like one
Scripts were created for a reason. They reduce risk and are intended in part to create consistency. They help new agents survive until they know what they are doing, but there is a point where script adherence becomes a substitute for thinking. And if the only thing we want from an agent is to read the approved line at the approved moment, then we should be honest and admit that we are designing work for automation.
A future-ready agent needs to understand why a policy exists, not just what it says. They need to know when a situation fits the standard answer and when it does not. They need to be able to make a decision when there is no perfect option, then explain that decision in a way the customer can understand.
That does not mean giving agents unlimited freedom, it means training judgment. There is a difference and that difference takes time and practice to be effective. Judgment can be coached, practiced and calibrated. But it cannot be developed if every training exercise has one clean answer and every QA form punishes agents for stepping outside the script.
If we want agents to think, we need to stop designing every part of the job to discourage thinking.
Empathy is not a line, it is a capability
Empathy has become one of those words everyone uses, but not everyone defines. In too many contact centers, empathy training means teaching agents to say something that sounds empathetic but is actually sympathy.
For example: "I understand how frustrating that must be." “if it were me, I’d also be upset.” (Insert eyeroll here)
Sometimes that line is fine. Sometimes it is empty. Sometimes it makes the customer more irritated because they know the agent has been trained to say it.
Real empathy is not the line. Real empathy is the ability to read the customer's emotional state, adjust your communication, and move the interaction forward without making the customer feel dismissed, minimized, or handled.
That is empathy in action. It is not sympathy. It is not just being nice. It is not letting the customer do whatever they want. It is a professional skill that helps an agent regulate the conversation so a resolution is even possible.
And I do believe it can be developed. Not everyone starts with the same natural level of empathy, and we should be honest about that too. But agents can learn how frustration works. They can learn what customers hear versus what the company thinks it said. They can learn how to acknowledge emotion without getting trapped in it. They can learn how to recover when the conversation starts going sideways.
The problem is that we often give this skill five minutes in training, then act like it should show up automatically on every call. It will not, not unless we train it like it matters.
A Language Score Is Not the Same as Live Communication Readiness.
This is especially important in offshore and nearshore BPO environments. We need to stop confusing language test scores with live communication readiness. An agent can pass a written assessment and still struggle on a live call. That does not mean the assessment was useless, but it does mean it was incomplete.
Live customer communication is different. The customer may speak quickly, or use idioms. They may jump around, interrupt, or explain the issue in a way that makes no sense until the agent asks the right clarifying question. They may be angry, embarrassed, confused, or impatient.
Grammar matters, but grammar is not the whole job. The agent has to listen for meaning, not just words. They need to compress a messy explanation into a clear summary. They need to repair the conversation when misunderstanding happens. They need to shift between formal and conversational language. They need to explain technical information in plain language without sounding robotic or condescending.
That is not a soft skill in the dismissive sense. That is a performance skill.In the future contact center, it may be one of the most valuable skills an agent has.
If we are serious about preparing agents for AI-era work, then communication has to be trained and measured as a live, applied capability. Not just as a grammar score or as pronunciation, or whether someone sounds neutral enough.
Can the customer understand them? Does the customer feel understood by them? Can they recover the conversation when it gets difficult? That is the real test.
AI Assists. The Agent Still Has to Think.
There is another mistake I think some companies will make. They will put AI tools in front of agents and assume that the agent will automatically become better. That is not how this works.
AI can help agents. It can summarize. It can suggest responses. It can surface knowledge. It can reduce search time. It can help newer agents move faster. An AI suggestion is not the same as the right answer, there should be an expectation that the agent still has to think.
They have to ask whether the suggestion actually fits this customer. They have to decide if the tone is right. They have to catch when the AI is technically accurate but emotionally tone deaf. They have to know when the system has misunderstood the context. This matters because one of the risks of AI in the contact center is that agents become passive repeaters of machine-generated language.
That would be a terrible outcome.
If we train agents to blindly accept AI output, we are not future-proofing them. We are reducing them to the last human step in an automated process. The real value is in knowing when to trust the AI, when to challenge it, when to edit it, and when to ignore it. That is the human layer, where judgment still lives.
The Customer Does Not Care Which Department Dropped the Ball.
The customer journey is becoming more fragmented, not less.
A customer may start in the app, move to the chatbot, search the help center, submit a ticket, receive an automated response, and then finally reach a human. By the time the agent gets involved, the interaction already has history. If the agent cannot see that history, that is a systems problem. If the agent can see it but does not know how to use it, that is a training problem. Either way, the customer experiences it as the same thing: "Why am I explaining this again?"
This is one of the great failures of customer experience, and honestly, it is embarrassing that we are still dealing with it in 2026. Customers should not have to restart the story every time the company changes channels, departments, or systems.
The future-ready agent needs to understand what happened before they entered the conversation. They need to know what the bot tried, what information was collected, where the handoff broke, and what the customer is likely feeling by the time they arrive.
That is systems thinking. It is not enough for agents to understand their part of the workflow. They need to understand the customer journey around their part of the workflow. Otherwise, we are just asking them to clean up broken experiences without giving them the context to do it well.
Resilience Is Real. Using It to Cover for Bad Operating Models Is Not.
I believe resilience matters. I also think companies sometimes use the word resilience as a polite way of asking people to tolerate bad operating models. To clarify, the answer is not to tell agents to toughen up while we keep giving them broken tools, unclear policies, poor coaching, unrealistic targets, and emotionally draining work. That is not resilience, at minimum it is ignorance and on the other end of the spectrum could be seen as neglect.
Real resilience is the ability to stay regulated during difficult interactions and recover after them. It is the ability to handle intensity without becoming defensive, cold, or careless. It is the ability to take the next call without dragging the last customer into it. It is a competency that can be trained, but it also has to be supported. We already have a blueprint with all of the necessary support around tough work processes like content moderation, Trust & Safety, etc.
If AI leaves humans with more of the hardest interactions, then companies have a responsibility to design support around that reality. Things like better coaching or debriefs, better-prepared team leaders, improved tools. If possible, better recognition of the emotional load of the work. Otherwise, we will keep losing good agents and then act surprised when attrition stays high.
Where This Training Actually Needs to Happen
If we are serious about building the agent AI cannot replace, then this work cannot live in a motivational speech or a one-hour module during onboarding, it has to happen across the workforce pipeline. Here are a few examples to consider:
Before hiring - Candidates need more than interview preparation. They need to practice the real skills of the job before they ever sit in front of a recruiter. They need to explain complex ideas clearly. They need to listen under pressure. They need to handle disagreement. They need to communicate in English in a way that is clear, adaptive, and human, especially if they are preparing for offshore or nearshore customer-facing roles. And they need to practice all of this under conditions that simulate real pressure, not calm rehearsal. The goal is not to learn what good communication looks like. It is to practice it often enough that it does not require conscious effort when the job is also demanding conscious effort for everything else.
During onboarding - BPOs need to keep teaching process, but they also need to add realistic practice. Not roleplays where the customer follows the script. Realistic scenarios with incomplete information, unclear policies, emotional customers, time pressure, and messy handoffs. The design principle matters here: practice conditions need to resemble real conditions. The goal is to make judgment and empathy more automatic under load, so those capabilities remain available when the environment is demanding everything else simultaneously.
Once agents are live -Coaching has to become more than QA scores and defect conversations. Team leaders need to be able to coach judgment, communication, emotional control, and AI collaboration. A scorecard can tell you what happened. A good coach helps the agent understand why it happened and what to do differently next time.
Outside the company -Agents need to know that they can develop these skills on their own too. Communication, leadership, AI fluency, confidence, and customer judgment are career assets. They do not only help someone survive the next version of the agent role. They help someone move beyond it.
This is where there is a real opportunity, especially in markets preparing talent for global BPO and customer experience work. It is important to also mention that training alone is not enough, there needs to be measures of success in the form of KPIs, and these KPIs are going to be new and in need of clear definitions, for the scale and how to improve performance to the metric.
We should not just be preparing people to get hired AND we should be preparing people to succeed after they are hired.
Those are not the same thing.
The Industry Has a Choice to Make.
AI will keep taking the easy work. That is not a prediction anyone should feel nervous making. It is already happening.
The best agents were never doing low-skill work. They were doing work the industry refused to measure properly. They listened to people who were confused, angry, or embarrassed. They translated company policy into human language. They made judgment calls in seconds that the policy document never anticipated. They held the relationship together when the process fell apart. AI is not making those capabilities less relevant. It is making them the job.
So if you are an agent, stop waiting for your employer to tell you this matters. Get serious about the skills that will still be valuable when the easy work is gone. Communication under pressure. Judgment without a script. Emotional control when the customer is at their worst. These are career assets. Build them deliberately. Measure them specifically and have continuous improvement techniques in place to see your workforce grow and master these new skills.
If you are a BPO or contact center leader, the ask is simpler and harder at the same time. Stop designing training programs for the version of the job AI is replacing. Start building the workforce that can do the work AI cannot.
The easy work is going away. What remains is harder, more human, and worth a lot more to the organizations that get this right.
Impactify builds tools for BPO talent acquisition and workforce development, including Evala.cx, a communication assessment platform that measures the real-world communication capabilities described in this article across voice and written channels for offshore and nearshore hiring environments.