PegaSystems CEO Criticizes LLM Token Pricing, Launches Token-Free AI Offering

PegaSystems CEO Alan Trefler has expressed his concerns about the high cost of tokens used by Large Language Models (LLMs). He notes that in recent weeks, the term 'tokenomics' has gained attention due to the unexpectedly high price of these tokens. This has led some companies to question their use and costs associated with them.

Trefler suggests that LLM providers may be building technology to drive token consumption, which could lead to a negative impact on users who are not aware of how tokens work. He argues that there is no added value in using tokens for reasoning at runtime, but rather a huge negative consequence: the inability to describe how things will work until after they happen.

PegaSystems has taken a different approach by 'burning' LLM tokens during design time and exercising them extensively. This allows users to run their applications thousands of times while only using the LLM for narrow tasks where translation or other specific functions are required. Trefler also points out that cheaper LLM models can be used in many cases, reducing costs.

The company has launched its Infinity 26 token-free agentic AI offering at the PegaWorld conference this week. This platform allows clients to design, build, and run their workflows without paying per token. The Predictable AI architecture shifts heavy AI reasoning to design time, making runtime agents fast, reliable, and cheaper to operate.

Trefler believes that customers are trying to figure out what's right in the face of rising skepticism and suspicion about LLM tokens. He notes that senior executives are questioning the value proposition and seeking education on how these technologies work. The emergence of tokenomics has added a new layer of complexity for enterprises to consider when making strategic decisions.

The concept of Reasoning Tokens, introduced by OpenAI and Anthropic, is not well understood by users, according to Trefler. These tokens are used internally by models to 'think' through complex tasks before producing a response. Despite some token price reductions on older models, new ones have been added with higher prices, making it difficult for users to track their costs.

Trefler criticizes the lack of transparency in LLM pricing and notes that users may be charged five to ten times more tokens than they initially expected due to internal processes like sub-agent interactions. He suggests that organizations should ask how these technologies can operate against them before adopting them, as seen with PegaSystems' cautious approach.

Trefler dismisses the idea of a 'SaaSpocalypse,' where LLM companies supposedly threaten to wipe out the software industry's Total Addressable Market (TAM). Instead, he believes that while some parts of software may be vulnerable to disruption, others will continue to thrive. He sees the need for predictable work management using intelligence and leveraging it as essential in the future.

Trefler also comments on the term 'tokens' being used instead of a more transparent label like 'words.' This has led to confusion among users about what they are actually paying for.