There’s been a debate surrounding autonomous AI agents for a while; it tends to swing between two extremes: euphoric techno-optimism promising bottom-line efficiency, and catastrophic anxiety over job displacement. But the reality of integrating AI agents into core workflows is far more nuanced – AI agents can offer extraordinary capabilities, but replacing human judgment at scale remains a high-stakes operational risk.
The good Ai : speed, scale, and efficiency
If deployed thoughtfully, AI agents act as an operational force multiplier.
- Instant throughput: AI agents can process millions of concurrent queries across dozens of languages without fatigue, drastically reducing initial resolution times.
- Elimination of routine friction: High-volume, rule-based tasks -such as password resets, order tracking, and standard billing queries – are handled seamlessly.
- Augmented human capability: AI tools can act as ‘co-pilots,’ surfacing relevant data, summarising complex cases, and suggesting responses to help human workers solve problems faster.
The bad Ai: the limits of empathy, context, and brand trust
Where pure AI agent replacement falls short is in handling nuance, complex edge cases, and emotional intelligence.
- The hallucination & compliance trap: AI models can confidently provide incorrect policy advice or misinterpret complex customer disputes, introducing financial and regulatory risk.
- Escalation friction: When an AI agent fails to resolve an issue, customers often re-enter the queue frustrated, forcing a human agent to pick up a fragmented, high-friction conversation.
- Brand erosion: Treating customer interactions strictly as a cost-cutting metric ignores the value of human connection in building long-term customer trust and loyalty.
The Klarna case study: The ‘boomerang’ effect
In 2024, Swedish fintech giant Klarna became the primary poster child for AI-driven workforce replacement. The company proudly announced that its OpenAI-powered AI assistant was doing the equivalent work of 700 full-time customer service agents, handling millions of conversations and saving tens of millions of dollars.
However, the aggressive push to eliminate the human layer soon revealed structural cracks:
Quality drop and customer friction: While simple queries were resolved instantly, complex cases, refunds, and policy edge cases stalled. Customer satisfaction declined on complex issues, and repeat contact rates spiked as customers returned to seek human help.
The course correction: By 2025, Klarna CEO Sebastian Siemiatkowski publicly acknowledged that prioritising cost cutting had led to lower quality, stating, “We went too far.”
The hybrid rehire model: Klarna quietly reversed course and resumed hiring human agents. Rather than returning to traditional call centers, they shifted toward a hybrid, remote model where human agents equipped with AI tools handle complex, high-value interactions, while AI absorbs the routine background volume.
The takeaway: automate volume, empower humans
The lesson from Klarna and similar corporate experiments is clear: AI agents should replace friction, not entire human relationships.
Organisations that succeed long-term do not treat AI as a quick headcount reduction tool. Instead, they build a hybrid operational layer where AI handles speed and scale, while human professionals are retained – and better equipped – to deliver judgment, empathy, and strategic resolution when it matters most.
Want to hear more about what’s happening with the world of finance? Come to financeSHOWCASE – the top speakers in the industry, with the leaders in tools for finance, what more could you want?
Register for your free tickets here
