Pegasystems CEO and founder Alan Trefler on AI agent ‘insanity’
Mainstream software program suppliers are pursuing a “philosophy of insanity” by persuading enterprises to deploy 1000’s of synthetic intelligence (AI) brokers, says Alan Trefler, the CEO and founding father of Pegasystems.
Trefler, recognized for his outspoken views, claims that enterprises are in peril of lining up issues for the long run by letting free en masse AI brokers with the power to make doubtlessly business-critical choices on their pc networks.
Trade watchers say Trefler has a degree in relation to mission-critical workflows within the regulated industries Pega provides, however that different software program suppliers are getting a grip on agentic AI.
Neil Ward-Dutton, analysis vice-president for agentic automation and AI applied sciences at analyst group IDC, advised Laptop Weekly that whereas there’s a race by suppliers to be seen to have AI capabilities, behind the scenes, different suppliers are introducing their very own guardrails for AI.
“I believe Pega realised fairly early on that you should use AI that can assist you create a deterministic system, like a Pega software. Now … all the opposite distributors are beginning to try this too, however Pega was one of many first,” he mentioned.
“Whenever you get behind the headlines and converse to ServiceNow or Salesforce or Microsoft, they may all discuss in their very own approach about this sort of hybrid method, the place sure, you should use AI, and you should use AI brokers, however truly you want to mix them with conventional, deterministic, extra rules-based approaches,” he mentioned.
The insanity of mass AI brokers
Trefler advised Laptop Weekly that mainstream enterprise software program suppliers are in peril of happening the fallacious observe by selling the concept that vital programs will be safely constructed on AI brokers.
“I believe they’ve already acknowledged they’re going to have 1000’s of brokers working, possibly over 10,000, and I believe that philosophy is insanity,” he mentioned.
The large enterprise software program suppliers are introducing “management towers” to permit enterprises to watch and handle their AI brokers.
However Trefler argues that AI brokers are too unpredictable for use at scale to run mission-critical functions in enterprises.
“[Software companies] have already acknowledged they’re going to have 1000’s of brokers working, possibly over 10,000, and I believe that philosophy is insanity”
Alan Trefler, Pegasystems
“You don’t need these disaggregated, disassociated initiatives attempting to run essential issues within the enterprise the place it would deal with clients in a different way in ways in which it shouldn’t,” he mentioned.
When AI goes fallacious, it could actually go fallacious in a spectacular style, as entrepreneurs who selected an AI-generated slogan for Starbucks in South Korea came upon, when their “Tank Day” marketing campaign led to nationwide riots and clients smashing Starbucks-branded cups, adopted by the resignation of its chief government.
Based by Trefler in 1983, Pegasystems is now on observe to develop into a $2bn income organisation.
Based mostly in Massachusetts within the US, the software program firm provides a low-code enterprise course of platform to a few of the world’s largest firms. It lists Deutsche Telekom, Lloyds Banking Group and Daimler Vans amongst its clients.
The AI bubble will burst
Trefler sees an enormous future for AI within the enterprise, however he’s cautious of its unpredictability and the hovering prices of AI tokens, and believes the present price of funding in datacentres to energy AI is unsustainable.
“It was very sobering to me to drive down Freeway 101 in California about two months in the past,” mentioned Trefler. “There was billboard after billboard of AI firms. I’m within the enterprise, and I didn’t know 80% of them.”
He predicts that when the AI bubble bursts, many of those firms are unlikely to outlive.
“I believe you’re going to get an actual flash within the pan from plenty of firms that will have had a good suggestion, however discover that there’s an enormous step between that and sustainable enterprise,” he mentioned.
Trefler has lived by tech bubbles earlier than. The fibre optic cable bubble burst in 2000 after firms laid big quantities of fibre optic cable that vastly exceeded demand.
There was “an inevitable monetary rebalancing”, with many fibre optic firms going out of enterprise. However over the subsequent 20 years, the surplus fibre was ultimately used.
If and when the AI bubble bursts, enterprises might want to re-evaluate what they need to use highly effective massive language fashions (LLMs) for, he suggests.
AI altering the character of the IT occupation
AI has already modified the character of the IT occupation. Trefler says software program platforms, equivalent to Pega’s Blueprint and Pega’s Infinity cloud platform, are more and more permitting folks with out IT abilities to design and construct enterprise functions.
Just a few years in the past, the identical duties would have required specialists in Pegasystems software program, who have been typically onerous to seek out and commanded prime salaries. At present, the design work will be carried out by folks with enterprise abilities somewhat than technical abilities.
Companies will nonetheless want technically skilled folks to combine complicated IT programs, however customary integrations are simply going to get achieved, in accordance with Trefler. “They’re already just about out of the field,” he mentioned.
Widespread use of AI means IT jobs will develop into “compressed”. There will likely be a necessity for fewer programmers, for instance, as massive language fashions churn out excessive volumes of code.
That has put IT professionals underneath stress to show their price by utilizing LLMs to “vibe code” pc code at excessive quantity.
“The difficulty is, we’ve all realized that having increasingly more code within the enterprise doesn’t make that enterprise extra dependable,” mentioned Trefler.
The necessity for predictability
Trefler describes himself as a lone “voice crying within the wilderness” and, going towards perceived knowledge, warns enterprises to assume twice about deploying agentic AI at scale.
He doesn’t see how letting 1000’s of AI brokers free to work together with clients, or make choices that have an effect on the enterprise, will be achieved safely or cost-effectively.
“What I’m saying is that enterprises that need to have a sure consistency of course of, consistency of soul, consistency of consequence, have to do it in another approach,” he mentioned.
AI is simply too unpredictable, warns Trefler. It might deal with clients in a different way in ways in which it shouldn’t, or it would by accident violate a regulation or a regulation.
“You don’t need these disaggregated, disassociated initiatives attempting to run essential issues within the enterprise,” he mentioned.
Over the previous two months, the financial influence of AI has develop into a rising problem, with CEOs beginning to fear concerning the rising value of AI tokens.
Trefler, solely half joking, compares present token worth hikes to a drug vendor successful over shoppers by providing free or low-cost medication, solely to ramp up the worth when they’re hooked.
The tendency to measure the influence of AI by the variety of AI tokens workers eat is misplaced, in Trefler’s view, with the one credible metric being whether or not AI helps companies lower your expenses or make cash.
The case for deterministic workflows
Neil Ward-Dutton, analysis vice-president for agentic automation and AI applied sciences at IDC, says Trefler is “proper on the cash” in relation to coping with the high-volume business-critical workflows, equivalent to checking the eligibility of individuals with insurance coverage claims, that Pega specialises in.
“It will be full insanity to attempt to change what Pega does with fleets of brokers working round, sort of guessing at what to do, attempting to make a fist of it. That will be an absolute catastrophe,” he added.
However he mentioned there are different, much less vital functions the place fleets of AI brokers can be simply high quality, equivalent to constructing advertising and marketing campaigns, translating promoting copy, or checking a web-based product catalogue is updated.
Software program brokers, equivalent to Claude Cowork and Microsoft Scout, sit on the desktop and will help folks schedule emails and work by duties comparatively simply.
He mentioned extra software program suppliers are coming round to an identical mixture of deterministic workflows and AI brokers, and are utilizing their very own applied sciences to execute agentic AI safely. This consists of utilizing their very own platforms to offer plug-in abilities for AI.
Salesforce, for instance, has launched AgentForce brokers, and is utilizing workflows and templates. A German provider, Cumunda, which competes with Pega, can also be utilizing a mix of deterministic workflows, with AI dynamic decision-making.
