Question 1

Why are AI agents so expensive to run?

Accepted Answer

AI agents are expensive because they require multiple inference calls per task (an agent completing a 10-step workflow might make 30-100 LLM calls), use reflexion loops that consume up to 50x the tokens of a single completion, and need expensive frontier models for reasoning-heavy steps. Unlike simple chatbot interactions, agents can't predict their compute costs in advance because the number of steps varies with task complexity and error correction needs.

Question 2

What is the failure rate of AI agents?

Accepted Answer

Current AI agents fail 50-75% of real-world tasks according to multiple benchmarks and production deployments. Enterprise environments typically require less than 1% error rates for automated processes, creating a massive gap between agent capabilities and enterprise requirements. Multi-agent systems face error amplification, where a 5% error rate per step compounds to a 17.2x higher failure rate across a 10-step workflow compared to single-step AI calls.

Question 3

Why did Klarna reverse its AI agent strategy?

Accepted Answer

Klarna initially claimed its AI agent handled two-thirds of customer service chats and replaced 700 human agents. The company later reversed course and began rehiring human agents after discovering quality degradation in complex customer interactions. CEO Sebastian Siemiatkowski acknowledged that AI could not fully replace humans for nuanced customer service. The reversal illustrates the gap between AI agent demo performance and production reliability at scale.

Question 4

What percentage of AI agent projects will be canceled?

Accepted Answer

Gartner projects that more than 40% of agentic AI projects will be canceled, scaled back, or restructured by 2027 due to escalating costs, unclear ROI, and implementation complexity. BCG research found that 60% of enterprises deploying AI broadly see no material business value. Initial cost projections for agentic AI implementations are typically off by a factor of 10x when accounting for error correction, human oversight, and infrastructure costs.

Question 5

How do AI agent costs compare to traditional software?

Accepted Answer

Traditional SaaS has near-zero marginal cost per transaction. AI agents have variable, unpredictable costs that scale with task complexity. A simple customer service interaction might cost $0.05 in inference, but a complex multi-step resolution with error correction can cost $5-50. OpenAI reportedly spends $2 for every $1 earned on inference across its product suite. Replit's margins swung to -14% when AI usage spiked, illustrating how agent-heavy products face margin volatility that traditional software never experienced.