Government CTO: 'I had a lightbulb moment with agentic AI'

AI could help shift the focus onto outcomes, says Defra’s Paul Mukherjee

Defra’s CTO believes AI may help government thinking move from processes to outcomes, while agents could assist in complex technical tasks like cloud migrations.

Despite the country's financial straits, this is a wonderful time to be a technologist in government, according to Paul Mukherjee, CTO of the UK Department for Environment, Food and Rural Affairs (Defra).

"The pressure to do things differently has never been as great, and it's only getting greater," Mukherjee said.

These days, when technologists say there might be a fundamentally different way of delivering services with improved outcomes, better quality and decreased costs, people listen.

"They are receptive to this message now, whereas I suspect they may not have been so receptive [in the past]," said Mukherjee, who joined Defra 18 months ago, speaking during a panel session at a ServiceNow public sector event in June.

Mukherjee said he was excited by the prospect that technologies such as AI could shift the focus of the civil service from processes to outcomes. Rather than streamlining existing delivery pipelines the real benefits will come from being able to completely redesign them, he stated.

"The big prize will be reimagining how we deliver an outcome, forgetting how we've done it historically."

For example, a hypothetical customer contact service could be delivered by default to 90% of the target customer audience using AI, with the 10% who do not want that or who have more complex needs being served another way. Currently, the hard cases are a blocker to change.

"We tend to use that 10% to say we can't do this, we need to change that thinking."

Not that becoming more outcomes-oriented will be easy. "There's a there's a mindset shift in at the top levels of the civil service that we're still working on achieving," Mukherjee said. “Landing something like that in a government department would be extraordinarily difficult.”

See also: Government tech plans at risk from lack of civil service talent, says former AI lead

There are also several ethical and operational questions that need to be resolved about the relationship between human beings and AI, including where responsibility lies for automated decisions.

Nevertheless, Mukherjee, whose previous roles included directorships in enterprise architecture in pharmaceutical and technology companies, said he was enthusiastic about the prospect of agentic AI being able to demonstrate new ways of achieving goals.

On the technical side he said he was already looking into how agents could automate complex processes.

"I had a lightbulb moment with agentic AI when I saw a demo of an agentic AI solution which could help us with our legacy organisation challenge, that automatically migrates applications from our legacy datacenters into the cloud, building cloud native applications. I was absolutely blown away, and we're actually exploring doing that."

Asked about how AI is likely to be used in government in the near future, Mukherjee conceded that in the short term it will be deployed for "cost takeout, because of where we are as a nation."

But that doesn't preclude a parallel move to more strategic thinking, he added. "It might be the only way to achieve that cost takeout is to go to outcomes-based thinking for more complex scenarios."

What’s more, new opportunities will emerge to deliver services that are not possible today, he went on. "As technology leaders within the public sector, it's incumbent on us to make sure that people are continually aware that it's not just about cost takeout. There are other outcomes we're able to deliver."