Pega targets token tax on agentic application development
Pegasystems Inc. (commonly known as Pega) used its PegaWorld conference in Las Vegas this week to explain how developers can design, build and run agentic workflows across Pega Infinity 26 without paying for tokens.
The trademarked™ Pega Predictable AI architecture comes with a promise of being able to shift the heavy AI reasoning to agentic function design time.
This way, runtime agents end up fast and cheaper to run.
In a world of escalating token costs and unreliable outcomes, Pega says that LLM providers are converting flat-rate subscriptions to more expensive token-metered pricing – all while running up expensive reasoning tokens behind the scenes.
The more complex the request, the more reasoning steps are required – and, actually, (claims the company) the more likely it will generate an inadequate and inconsistent answer.
Design-time reasoning
Pega applies AI reasoning at design time, when (the company argues) creative power delivers the most value for reimagining outdated processes and systems.
“With Pega Blueprint AI and the new Pega Infinity Studio, Pega’s design agents help teams design and build the optimal agentic workflows for their mission-critical business processes. Common examples include servicing a customer request, approving a loan, underwriting a claim, or optimizing a patient experience,” said the company.
Once the workflows are designed and deployed, Pega shifts to a lighter-weight semantic mode of AI better suited for runtime, when agents are called on to process millions of user requests efficiently and consistently.
Instead of re-reasoning each new workflow, agents use a lightweight AI query to understand the user intent, find the best Pega workflow for the job and then follow it step-by-step to complete the work.
NOTE: If a specific step needs deeper LLM use (e.g., to parse a document or summarize previous interactions), the step provides specific and bounded instructions to ensure predictability.
Tallying up token calculations
To help enterprises quantify the benefits of this approach, Pega has introduced the AI Token Cost Calculator. A tool that estimate possible savings by comparing Pega AI with token-metered alternatives based on users’ workflow volumes.
“Enterprises are quickly waking up to the fact that tokenmaxxing is ridiculous: it can only lead to unsustainable costs and unpredictable results,” said Alan Trefler, founder and CEO, Pega. “AI best creates value when it delivers reliable outcomes at scale. That’s why we don’t charge clients based on how many tokens they use, but by the meaningful work they accomplish. Combined with an architecture built for governed execution, Pega now gives organisations unrivalled freedom to use AI agents.”
Pega claims that “many clients” can realise a savings of more than 20x, depending on workflow complexity and scale. Pega’s outcomes-based approach charges per completed ‘case’ – a task executed from start to finish – not per seat or per token. For example, when a customer uses an AI agent to change an existing order, that completed interaction is recorded as a single case.
