Agentuna Agentuna

Real example · Customer service & WhatsApp

Klarna replaced support agents with AI, then hired humans back

Done withChatGPT
Free tier From $8/mo

What they did

Klarna launched an AI customer-service assistant, froze hiring for over a year and said in 2024 it did the work of 700 agents. In 2025 its CEO said the company would recruit human support agents again.

How it went

Entrepreneur reports the chatbot handled 75% of chats (about 2.3 million) within a month of launch and headcount fell 22% to 3,500, mostly through attrition. The CEO later said the AI was cheaper but gave lower-quality support.

Worth knowing

Account of a Bloomberg interview as relayed by Entrepreneur, with no detail on which cases failed. The 700-agent figure is Klarna's own claim, and the company says it is not abandoning AI.

Try it yourself with ChatGPT

Replace anything in [brackets] with your details.

Prompt to paste
Design a support triage policy for [business]. List the 10 most common customer questions and mark each as: the AI can fully answer, the AI drafts and a human approves, or human only.
For the human-only cases, write the handoff message and list what the AI must collect first. Add a rule that customers can always ask for a person.

Read the original

Klarna CEO Reverses Course By Hiring More Humans, Not AI (Entrepreneur) ↗
entrepreneur.com · · News report

Attributed to the source linked below; not our own test.

Share Share on X Share on WhatsApp

Phoned patients about abnormal vitals with a voice AI agent ElevenLabs

Phoned patients about abnormal vitals with a voice AI agent

Cadence, a remote-care company for chronic disease, built a voice agent that calls a patient when a vital sign is out of range, runs a symptom check under written clinical protocols and warm-transfers to a nurse when needed.

Cadence (remote care for chronic disease) · elevenlabs.io · · 1 account

Run a WhatsApp agent that matches community members with each other Make

Run a WhatsApp agent that matches community members with each other

OPUS, a community platform for early-stage founders, built two Make AI agents and about 49 workflows. One gathers member data from forms, call transcripts and voice notes; the other, called "operator", answers members on WhatsApp and makes introductions when both sides agree.

OPUS (via Make case study) · make.com · · 1 account

Answer routine health-platform questions on WhatsApp in Arabic WATI

Answer routine health-platform questions on WhatsApp in Arabic

Altibbi, an Arab-world telehealth and Arabic medical-content provider, moved its WhatsApp conversations to WATI, using chatbot automation for routine and subscription questions plus bulk campaigns and live reporting.

Altibbi (via WATI case study) · wati.io · · 1 account

n8n

Automate blocked-payment support tickets end to end

Koralplay used n8n to handle blocked-payment tickets the way a level-2 agent did: log into the back office with multi-factor authentication, look up the transaction, check regulatory status, update the ticket in Notion and message the customer with the answer.

Koralplay (via n8n case study) · n8n.io · · 1 account

Unifonic

Ran Ramadan donation campaigns over WhatsApp

ZMZM, a non-profit health charity in the Makkah Region, used Unifonic's WhatsApp messaging as its main outbound channel for Ramadan, with one-tap donation flows and a chatbot that answers questions and hands off to human agents.

ZMZM (health charity, Makkah Region) · unifonic.com · · 1 account

Wired WhatsApp Business to an n8n AI agent that answers customers n8n

Wired WhatsApp Business to an n8n AI agent that answers customers

Aziz Basha builds an n8n workflow in which a WhatsApp Business message reaches an AI agent node backed by an OpenAI chat model, which analyses the message and sends the reply. He walks through the Meta developer setup and shares the workflow files free.

Aziz Basha (YouTube) · youtube.com · · 1 account