Klarna CEO: AI agents will kill SaaS switching costs
Klarna CEO Sebastian Siemiatkowski predicted that the next major disruption in software will be the reduction of "switching costs" between platforms, driven by AI agents that can easily migrate data. He argued on the *20VC* podcast that as the cost of creating software approaches zero, revenue multiples could fall from over 20x to as low as 1-2x. Siemiatkowski envisions a future of modular, "Lego piece" software where AI acts as the primary integration and orchestration layer.
- A core challenge for AI labs is "model alignment," ensuring AI behavior is helpful and harmless; one method, Constitutional AI, developed by Anthropic, trains models using a set of principles (a "constitution") to self-critique their outputs, reducing the reliance on large-scale human feedback. Reinforcement Learning from Human Feedback (RLHF) is another key technique where human labelers rank or compare AI-generated responses to create a "reward model" that guides the AI's future outputs. - The rise of agentic AI creates new evaluation challenges beyond traditional text-quality metrics. Specialized benchmarks are now used to test agent performance, such as WebArena for web navigation tasks, GAIA for general reasoning, and ToolBench for evaluating the correct use of external tools and APIs. Early GPT-4 agents had a 14% success rate on the WebArena benchmark, which later improved to around 60% with better agent design, showing the steep learning curve and the need for high-quality evaluation data. - AI labs are increasingly turning to synthetic data—artificially generated information that mimics real-world data—to solve for data scarcity and privacy issues. This approach can reduce data acquisition costs by 60-80%, and Gartner projects that by 2030, 60% of all data used for AI development will be synthetic. However, the highest-performing models often use a hybrid approach, combining real and synthetic data for training. - The data annotation market is shifting from large-scale, simple labeling to requiring high-quality, expert feedback, especially for frontier models. AI labs at OpenAI and Anthropic are increasingly sourcing feedback from domain experts in fields like law, science, and software engineering to evaluate nuanced tasks like code generation and complex reasoning. - The go-to-market (GTM) strategy for B2B AI startups selling to technical buyers emphasizes focusing on outcomes over technology. Instead of describing the underlying model ("LLM-powered root cause analysis"), successful messaging highlights quantifiable value, such as "cut debugging time by 40%." - The fundraising environment for AI infrastructure has seen explosive growth, with AI-focused companies capturing nearly 50% of all global venture funding in 2025, up from 34% in 2024. Foundation model developers like OpenAI and Anthropic alone raised $80 billion in 2025, accounting for 40% of global AI funding. - While AI is projected to displace millions of jobs, it is also expected to create new roles. A World Economic Forum report estimated that AI would displace 75 million jobs globally by 2025 but create 133 million new ones. Another forecast from Goldman Sachs suggests AI could replace the equivalent of 300 million full-time jobs while increasing global GDP by 7%.
