Agentic AI customer service in 2026: What it really means for BPOs and their clients
- Agentic AI customer service takes autonomous action on the customer’s behalf — resolving tickets, processing refunds, and executing multi-step workflows end to end.
- The promise is real: 24/7 resolution, lower cost per contact, and headcount reduction at scale.
- Full headcount replacement is unrealistic on a one-year horizon. The next three to five years favour hybrid models with humans supervising AI on complex work.
- The winning move for BPOs and CX buyers is to lean in early: supervise AI rather than fear it, scope pilots by intent, and design outcome-based pricing models before competitors do.
Silicon Valley’s pitch deck for the post-BPO future is simple: agentic AI will wipe out the human element of customer service, and operators who don’t pivot will go the way of Kodak. Public BPO share prices have already taken the hit.
But for operators deploying agentic AI customer service today, the reality is messier than the slide deck.
In the 589th episode of the Outsource Accelerator Podcast, Simone Bartlett, co-founder and CEO of Africa Operations at Hugo, is direct about both the threat and the trap. Hugo is a 4,500-person outsourcing firm across Nigeria, South Africa and the US.
The technology is unstoppable, in her view. The timeline is not.
What is agentic AI customer service?
Agentic AI customer service refers to AI systems that don’t just answer questions. Instead, they take actions on the customer’s behalf.
Unlike chatbots or IVR, agentic AI can navigate workflows, trigger refunds, update accounts, and execute multi-step resolutions end to end.
Gartner frames it as a new paradigm where AI systems “possess the capability to act autonomously to complete tasks.”
For Simone, it’s the live conversation: “In CX, the big talk now is agentic.”
The promise and honest trade-off of going agentic
The pitch is compelling: 24/7 resolution, lower cost per contact, no seat-based scaling pain. Gartner expects 40% of enterprise applications to include task-specific AI agents by the end of 2026, up from less than 5% in 2025.
The catch is what that does to BPO headcount. Simone doesn’t dress it up:
“It’s difficult to go tell one of your best clients, ‘Have you started thinking about agentic and how we can layer into our way of working?’
Because it will mean headcount reduction. When done well, it will mean headcount reduction.”
Operators willing to have that conversation early shape the work that remains. Those who hide get cut out of it altogether.
Why full replacement isn’t happening yet in customer service
Despite the projections, end-to-end agentic AI customer service isn’t a present-day reality at scale. McKinsey’s 2026 AI Trust research flags the new risk: organisations must contend not just with systems saying the wrong thing but doing the wrong thing.
Simone puts it sharper:
“Simple mistakes will shake trust. If we say, ‘Let’s go fully agentic,’ and we don’t have humans, and then something goes wrong, [you’d think], ‘I should have humans just overseeing or checking it.'”
Her analogy: “Even if the teenager’s really smart, the 40-year-old has wisdom from having seen it many times.”
Her working window before AI handles routine work unassisted is three to five years.
The hybrid model for agentic AI customer service
Hugo’s bet on agentic AI customer service is not on the AI itself, but on the layer above it.
“The one thing that has never been wiped out in the whole wide world are supervisors. How do we become people that help AI triage?”
Lower-complexity contacts get absorbed by agentic AI; mid- and high-complexity work moves to humans who supervise, evaluate, and step in.
A 2026-realistic split looks like this:
| Customer Service Task | Agentic AI Handles | Human Supervises / Owns |
|---|---|---|
| Password resets, order status, balance enquiries | ✓ | Spot-check QA |
| Refunds, returns, policy lookups | ✓ (within rules) | Escalations + edge cases |
| Multi-system account changes | Partial | Final action + audit |
| Complaints, churn risk, complex billing | Triage only | ✓ |
| VIP / enterprise account handling | Assist | ✓ |
| Model evaluation & guardrail testing | — | ✓ |
How to approach agentic AI customer service
Translating the hybrid model into action takes a deliberate playbook. Four moves separate the operators leaning in from those waiting it out.
1. Don’t hide
The Kodak risk for BPOs is not too much disruption — it’s pretending it isn’t coming.
Sione says “If I don’t show courage in confronting it, the door is open for others that are willing to just come and exist in that paradigm.”
In practice, that means raising agentic AI with clients before they raise it themselves. The operator who opens the conversation keeps the relationship.
2. Reframe the question
Move from “where can AI supplement our service?” to “where can our service supplement your AI?”
Buyers in 2026 are increasingly agentic-first. They’ve already piloted internally and want partners who fit around their stack.
Lead with that framing and you position yourself as a complement to the buyer’s AI roadmap, not a casualty of it.
3. Pilot on lower-stakes flows first
McKinsey’s January 2026 analysis shows scaled agentic adoption is concentrated where errors are cheap — IT, software engineering, knowledge management.
Customer service follows the same logic. Scope agentic AI by intent: start with high-volume, low-risk flows like password resets and policy lookups, prove reliability, then expand the agent’s remit.
4. Rethink the pricing model
As agentic AI cannibalises billable hours, the per-FTE model erodes. The alternative is outcome-based pricing — billing per resolution or per outcome delivered — but no one has fully solved it.
Simone conceded Hugo is still working it out: “If we create value, how can we capture some of it? That’s always a discussion.”
The BPOs that design new commercial models early will set the benchmark.
FAQs
Is agentic AI customer service already replacing call centre jobs?
Some routine contacts, yes — but replacement claims outpace what’s shipped. Gartner’s 2026 Hype Cycle places agentic AI at the Peak of Inflated Expectations, with only 17% of organisations having deployed AI agents to date.
How fast will agentic AI customer service mature?
Gartner projects 80% autonomous resolution of common issues by 2029. Most operators expect a three-to-five-year transition window where hybrid models dominate.
What roles will survive in BPOs?
Supervisors, escalation handlers, model evaluators, and operators of high-complexity workflows. The lower-complexity tier shrinks first.
Should BPO clients bring everything back in-house?
Not necessarily. Building, governing, and supervising agentic AI is itself becoming an outsourceable capability.
Key takeaways
- Agentic AI customer service is real and accelerating — but full replacement is a three-to-five-year story, not a 12-month one.
- The dominant 2026 operating model is hybrid: AI handles routine resolution, humans supervise and own complex contacts.
- For BPOs and CX buyers, the winning move is to lean in early — restructure pricing, build supervision capability, and have honest headcount conversations now.
- Operators thinking like Hugo — wisdom above intelligence, supervision above replacement — will set the benchmark for what good agentic AI customer service looks like.
Independent
