Interview: Pegasystems’ Don Schuerman on the way to hold the lid on skyrocketing AI prices
The rising price of utilizing giant language fashions (LLMs) is now giving enterprises pause for thought. As synthetic intelligence (AI) fashions have change into extra refined, queries are costing companies considerably extra in “tokens” and, in some instances, ratcheting up disastrously giant payments.
And their inherent variability means it’s not doable to foretell how a lot every activity will price. The identical immediate at some point may produce an prompt response, however an virtually similar immediate on one other day may take 5 minutes and burn via 10% of your month-to-month token funds, says Don Schuerman, chief know-how officer (CTO) at Pegasystems.
Enterprises have incentivised staff to maximise their use of AI with out totally contemplating the advantages that it produces for the organisation. Such “tokenmaxxing” has left corporations with sudden payments.
Final month, for instance, Axios reported {that a} single unnamed enterprise spent over £500m on AI tokens for Anthropic’s Claud AI platform in a single month after failing to place a cap on its staff’ IT use.
And in April, Uber’s CTO disclosed that the minicab and supply service had burned via its whole funds for Claude Code for 2026 within the first three months of the yr.
AI token price rising
Business pressures to fund AI datacentres and growing power prices have led AI suppliers to lift their costs in current months, main organisations to query the place the worth of spending on AI lies.
In some instances, Schuerman says corporations have changed individuals with AI solely to understand that AI is costing them greater than the individuals they changed.
Final week, Mary Daly, president and CEO of the Federal Reserve Financial institution of San Francisco, put it succinctly in an interview with Bloomberg. “Productiveness progress is in every single place besides within the information,” she stated.
“What’s occurred is the fashions have gotten extra refined,” Schuerman tells Laptop Weekly. “The mannequin causes with itself, typically it dispatches different brokers to do different issues, and because it does that, it’s constantly working the token meter.”
Enterprises waking as much as AI prices
He argues that enterprises are waking as much as the truth that the price of AI doesn’t enhance linearly with the variety of calculations the mannequin makes. “Each step is including quadratically to the price of the method,” he says.
When you use AI to work out a enterprise course of, step one would possibly take 500 tokens, however the context of step one will have to be carried over, so the second step would require 1,000 tokens. The third step will use 1,500 tokens, and so forth, says Schuerman.
As calculations change into extra complicated and require extra context, not solely does the associated fee enhance, however the threat of AI hallucinating or behaving unpredictably additionally will increase.
“The absolute best use of AI is to assist me get that repeatable course of proper – assist me outline it, assist me design it, assist me guarantee it follows finest practices. And it seems I don’t want a lot AI for that”
Don Schuerman, Pegasystems
Pegasystems’ reply to preserving the prices of AI underneath management is to make use of the know-how in a extra strategic means.
Pega provides Fortune 500 corporations with low-code platforms to automate their enterprise processes and handle relationships with their clients.
This week, the corporate introduced that it will cost its clients for enterprise outcomes, somewhat than charging them for what number of AI tokens they use. Schuerman argues that at the least 60% to 70% of the high-volume mission-critical processes enterprises want may be automated utilizing rules-based approaches.
“The absolute best use of AI is to assist me get that repeatable course of proper – assist me outline it, assist me design it, assist me guarantee it follows finest practices,” he says. “And it seems I don’t want a lot AI for that.”
Designing workflows
Pega’s Blueprint software program, for instance, makes use of AI to assist individuals design automated workflows for his or her organisations. As a result of AI does the majority of its work on the level of design, AI brokers don’t must rethink the method from scratch every time the method runs.
The workflows can, nonetheless, name AI brokers to execute particular duties – summarising a doc or in search of enter from a human, for instance – giving enterprises the flexibility to make use of AI reasoning in a managed means.
Pega has now made its automated workflows appropriate with the open supply Mannequin Context Protocol (MCP). This implies corporations can provide AI brokers constructed on different platforms – resembling Anthropic Claude, Google Gemini, OpenAI, AWS AgentCore, and others – entry to Pega’s processes.
Schuerman says that for Pega, including MCP was a “comparatively gentle carry”. Pega is agnostic about how organisations entry its workflows. Organisations can already entry Pega workflows via different software program, resembling Salesforce, for instance, with out having to make use of Pega as a entrance finish.
However for companies, the change is big, he says. They will take their present AI brokers and use them to observe Pega’s workflows.
“As soon as I’ve related within the Pega MCP, that agent goes to observe the foundations, it’s going to do it with out extreme reasoning [by making] that comparatively easy name to search out the suitable workflow and discovering the ability,” says Schuerman.
From banks to pizza
Firms, together with the Dutch financial institution Rabobank, are utilizing MCP calls to transform chatbots into clever brokers that full duties for patrons, resembling checking an account steadiness or doubtlessly making a fee.
Blueprint has opened the door for extra corporations to change into Pega clients, extra shortly. Previously, it might need taken a number of years for Pega to develop prototype workflows and for corporations to determine whether or not to make use of Pega software program.
“It typically took an extended set of conversations and actually expert salespeople to assist the client perceive how their enterprise downside match into our platform,” he says.
With Blueprint, the software program can take an outline of what they do as a enterprise, scan materials from the corporate web site, and inside minutes design workflows that make sense to the enterprise.
Schuerman gained’t put any numbers on it, however says Pega has made a acutely aware determination to promote its platform to a broader vary of corporations, partly straight and partly working via enterprise companions.
Beforehand, Pega specialised in supplying its platforms to extremely regulated industries resembling banks, insurance coverage corporations and healthcare. However now it’s seeing alternatives from corporations like Papa John’s, the pizza chain, which was seen among the many attendees of Pega’s annual convention.
AI can’t create the extraordinary
Schuerman is adamant that AI isn’t going to switch human creativity as enterprises change into extra automated.
In contrast to individuals, AI doesn’t create issues which can be extraordinary or present real creativity. “It simply averages all the things else that it has learn previously,” he says. “What we wish to do is regularly compress the time from an thought to that concept making a significant change in how my staff work and the way clients interact with us.”
That leaves individuals free to focus extra on technique, concepts and creativity. There may be an excessive amount of deal with the following model of ChatGPT or Claude, when final yr’s model was simply tremendous, in line with Schuerman.
“The arduous work is in turning all this AI potential into realities for individuals, and that’s the place I believe the attention-grabbing work is,” he says.
