In January 2024, a customer trying to track a missing parcel ended up with a poem instead. After the customer shared screenshots online of how he had got the chatbot to swear and criticise the company, DPD switched off the AI element of its chat. The company said the problem followed a system update.

It made headlines because it was funny. It’s worth remembering because of what came next.

Three headlines, one pattern

A few weeks later came a case that wasn’t funny. A customer asked Air Canada’s chatbot about bereavement fares and was told he could claim the discount after travelling. The airline’s published policy said otherwise. When he sought the refund, Air Canada argued it was not responsible for what its chatbot had said. A British Columbia tribunal disagreed and ordered the airline to compensate him.

In April 2025, a software company’s support bot told a customer about a one-device-per-subscription policy that did not exist. Users said online that they would cancel, and the company confirmed the bot’s answer was wrong.

Look at what these three have in common. None of those systems did anything. They only talked, and words alone were enough to cost money, trust and, in one case, a legal ruling. In each case, the company was left owning the consequence.

Now give the system hands

Everything above involved software that answers. The next generation is built to act. Gartner’s analysts describe the shift this way: earlier GenAI tools help people with information, while agentic AI is designed to resolve service requests on a customer’s behalf.

Imagine a system that can issue a refund, change a booking or cancel a subscription. That changes what a mistake looks like. A wrong answer becomes a wrong outcome.

The Klarna experiment

Few companies tested this in public as boldly as Klarna. In February 2024, it announced that its AI assistant had handled two-thirds of its customer service chats in its first month, which it said was equivalent to the work of 700 full-time agents. Those were Klarna’s own figures.

In May 2025, Klarna’s CEO told Bloomberg that cost had become too dominant a factor in how the service was organised, and that quality had suffered as a result. He said customers should always be able to reach a human if they want one.

Both chapters can be true. The efficiency claims and the quality concern came from the same company. The question isn’t whether AI can handle volume. It’s what is being measured while it does.

What the analysts say

Analyst outlooks for agentic AI are ambitious, but even they urge caution. Gartner has warned that more than 40% of agentic AI projects are likely to be cancelled by the end of 2027, citing rising costs, unclear business value and inadequate risk controls. It has also cautioned about “agent washing,” where existing chatbots and assistants are rebranded as agentic. Its advice is to pursue agentic AI where it delivers clear value or return on investment.

What customers say

Gartner’s customer survey in early 2026 found that people are becoming more comfortable with AI resolving service issues, yet 87% said it is essential to be able to reach a human agent when a company uses GenAI. According to Gartner, customers push back when AI becomes a barrier to reaching a person, and among those unwilling to engage with AI, the most common thing that might change their mind was the option to switch to a human.

In other words, customers seem to be asking for an exit more than for the end of the technology.

A region building rather than debating

In the UAE, the ambition is explicit. Abu Dhabi’s government digital strategy for 2025–2027 aims to make it the world’s first fully AI-native government across its digital services by 2027. It is a government treating AI as infrastructure rather than a pilot, and it will produce real deployments to learn from.

Five questions these stories raise

These are my takeaways from the examples above:

1. Does it answer, or does it act? The risk changes completely.
2. Who owns the outcome? DPD, Air Canada and the software company each ended up owning it.
3. Where is the human exit? Klarna’s CEO and Gartner’s 2026 survey point the same way.
4. What are we measuring? Klarna’s early claims centred on speed and volume.
5. Is the use case worth it? Gartner recommends pursuing agentic AI only where it delivers clear value.

Where these conversations continue

These are the questions CX leaders in the region are working through, and many of them will be in one room on 19 January 2027 in Dubai at Agentic CX Unplugged 2027.

The agenda opens with a keynote titled “From Chatbots to Autonomous Agents: The New CX Operating Model.” It is followed by a panel on whether agentic AI is ready for the enterprise, covering trust, governance and risk, ROI and workforce readiness, and by case studies from banking and financial services and from retail and e-commerce.

Around 100 CX leaders are expected. The event is curated by Scribe Minds & Media and held in association with the Middle East – UAE Agentic AI Forum (20–21 January 2027).

Join us in Dubai and bring your toughest questions: https://www.agenticaiforum.net/agentic-cx/