Appian’s CEO predicted an “agent meltdown” later this year, as the wheels come off earlier iterations of the technology that had been launched without the proper guardrails.

Appian cofounder Matt Calkins made the prediction as he unwrapped the firm’s own approach to agentic AI at its annual customer event in Denver.

Agentic AI is certainly riding the hype wave at present. NVIDIA's Jensen Huang recently predicted billions of "digital workers" labouring away alongside humans before too long.

Calkins likened the amount of investment being poured into AI to the US railway boom of the 1800s. “It paid off,” he said. But also was a monopoly and spawned other monopolies. AI, by comparison was open, or at least “almost free… there's many choices, different models. They're all roughly equivalent in power.”

But the railroads also had a clear application – transportation of goods and people - he said. AI by comparison, didn’t necessarily have a clear killer app. “It’s an open question.” And it was clouded by hype. “You would think that AI was for thinking around corners and inventing leaps of creativity”

This was not the case, he said. “The truth is, what AI is good for? AI is for doing regular jobs with superhuman efficiency.”

So, he said, leaders needed to look around their organisations, and “Look at the most important thing and look at the thing you do most frequently….that’s where you put AI.”

“AI is supposed to get its hands dirty doing the work, the core work, the essential, mission, critical, high volume, important work.”

Which is what you’d expect from Appian, given it styles itself as the process company. But it is embedding agents in these processes that would deliver the outcomes leaders are starting to demand from their vast investments in AI, he said.

The company already includes some agents in its offering. But in Denver, it took the wraps off Agent Studio which it said will make agents “more powerful and governable than ever.

Gregg Aldana, VP and head of global solutions consulting, demonstrating the technology to developers at the event, claimed it would allow companies to automate “dynamic, open ended tasks” that don’t “fit neatly into a process model.”

Appian’s approach meant “You give the agent a goal of plain language, it gets access to all the tools, all the records, all the rules, all the documents, and then it determines the best path forward to achieve that goal.”

Aldana said traditional RPA was tricky and rigid. “That's why we're evolving from rigid bots to intelligent agents and automation that can adapt on the fly.”

Appian execs said the company had deliberately held back on its agentic technology because its customers were “very serious organisations” and it wanted to ensure that it had the right guardrails in place.

In his keynote, Calkins said that the agents its customers produced would be “safe” and auditable. “When you ask an agent to act in processes you are constraining its behaviour to only those actions you feel would be constructive, albeit very powerful, long reaching actions.”

He predicted “A little bit of an agent meltdown later this year from some vendors who have agents but do not constrain them to constructive, scripted, guardrailed actions, just to guess.”

Aldana said companies were already exploiting Appian's existing AI technology. He cited the example of health service provider Acclaim Autism which had used Appian technology to reduce patient onboarding from six months to ten days.

The company also highlighted its Data Fabric, which can pull together data from multiple systems. Aldana said “I know there's a lot of companies out there saying they have a data fabric. It's a bunch of PR. This is the real patented Data Fabric.”

He said this removed the need to stitch together multiple data sources. “You can now seamlessly connect and work with data from anywhere, Salesforce, SAP, legacy systems, spreadsheets, documents, you name it, we can connect to it.

Aldana said the Data Fabric created “an enterprise wide virtual database with real time read and write access across all of your underlying systems.”

He said the next release, next month, would introduce “documents as records”, which would mean record types could be used to store and manage documents directly in Data Fabric.

"In the past, documents were harder to organize, harder to secure, time consuming to work. Now mapping this documents records is as simple as just a few clicks.” He added documents can be related to other data.

Join peers following The Stack on LinkedIn

The link has been copied!