OTTAWA — The federal government is definitely going to buy a new government-wide artificial intelligence system, and will seek the AI industry’s help to determine what to get and why, Prime Minister Mark Carney’s office says.

The government will have to fit it together with the many AI tools public servants are already using and apply the tool to the government’s many varied data stores, yet avoid the pitfall—flagged by its own top AI experts—of using AI for its own sake.

In their Nov. 4 budget, the Liberals announced plans to develop “a made-in-Canada AI tool that can be deployed across the federal government.”

The government’s central technology agency, Shared Services Canada, will lead the effort, with help from the Department of National Defence and the Communications Security Establishment, the budget said. “By supporting innovative research to strengthen public services, this work will protect our digital sovereignty, keep government data and information safe in Canada and create opportunities for the Canadian technology sector.”

Canadian AI vendors will help the government figure out what practical uses the tool can be put to, one of the prime minister’s spokespeople, Audrey Champoux, told The Logic in an email.

“The tool may not even exist yet, depending on what the need will be. Or maybe part of the tool exists, and we’ll work with AI companies to refine it to meet the demands [Shared Services] identified,” she wrote.

The federal government has a deepening relationship with Canadian AI company Cohere, having helped finance Cohere’s computing needs and begun trying out its offerings under a memorandum of understanding signed this fall. Cohere has a platform called North that promises to make it easy to create AI agents that can complete digital tasks, going beyond large language models’ synthesizing of textual information.

Cohere might help determine the government’s AI needs but that hasn’t been decided, Champoux told The Logic.

The Liberals ran in the election earlier this year on a promise to “look at every new dollar being spent through the lens of how AI and technology can improve service and reduce costs.”

The budget listed some ways federal departments plan to use AI to do that, in general terms. Shared Services itself wants to use it to automate responses to common tech support requests. Justice Canada wants to “streamline routine tasks, enhance decision-making and free employees to focus on higher-value strategic work.” Transport Canada has similar non-specific intentions to “optimize back-office activities.”

A Shared Services Canada spokesperson told The Logic the agency is already “leveraging both in-house and commercially available AI solutions to enable the government of Canada with AI tools that increase efficiency and productivity.”

It will continue ramping up use of the tech to “collect and analyze feedback on AI products to refine solutions and expand AI operations to additional users and organizations,” spokesperson Jeremy César wrote in an email last week. 

As for the new tool: “Additional details about the AI tool and its implementation will be shared once the budget is passed and further information becomes available,” César wrote.

The government is already using numerous AI tools, the federal chief information officer, Dominic Rochon, said on stage at an Ottawa technology conference last month: “I refer to AI adoption in the federal government as 1,000 flowers blooming, and I don’t mean that in any joking way,” he said.

Immigration and social benefits applications are being triaged and processed faster with AI, he told attendees at the Tech7 conference put on by Technation, which represents IT companies that do a lot of government business. Fisheries and Oceans is using AI to spot marine mammals in satellite and drone images. The RCMP is using AI in human-trafficking and child-exploitation investigations. 

Shared Services itself has a chatbot for government employees to use when otherwise they might turn to ChatGPT. CANChat is trained to “know” things like Canada’s April 30 tax deadline and designed to not train on user inputs, for security.

Rochon said there are so many AI things already going on that even he doesn’t know what they all are.

“We’re looking at, first and foremost, compiling a registry so that we understand which flowers are actually blooming—trying to measure whether or not they’re achieving what they’re setting out to achieve in terms of productivity,” he said.

One of the federal government’s senior-most AI mavens, Mark Schaan, warned at the same conference about rushing to use AI for the sake of using AI.

He cited the hypothetical scenario of a C-suite executive who went to a conference and got excited.

“Then they came back into the boardroom or the management room and said, ‘We’ve got to do it, like, now. Now is our time. Everybody’s doing it,’” Schaan said.

A longtime public servant who has spent most of his career at Industry Canada working on policies for innovative industries, Schaan did a 16-month stint as deputy secretary to the cabinet for artificial intelligence in 2024 and 2025, and is now the top public official under AI Minister Evan Solomon.

Good AI deployments require deep thinking, he said.

“The remarkable opportunity of AI is to take a business process that potentially is fraught and full of inefficiencies and revisit its fundamental first priorities and objectives,” Schaan said. An organization has to ask why it does what it does and how it can best serve its customers, clients or citizens. “And then you get to reimagine—where does tech fit into that, and what are the tools that are going to enable us?”

Preparation—long, slow, slogging preparation—is also key to making AI tools work, he said. Schaan said he’d been pleased to learn from a big financial-services company that its people spent two years organizing its stores of data so that AI models could use them effectively.

“I probably shouldn’t get overjoyed at the notion that an organization had a two-year data-cleanup exercise, but it is the truth of how we’re actually going to get to benefit,” he said.

Data quality is a long-standing challenge for the government, which has numerous digital systems that it developed separately—over decades, in some cases. It’s near the end of its second multi-year data strategy trying to reckon with the problem, and still has a lot of work to do. Just in the relatively narrow sphere of tracking what the government buys, disparate systems lead to “issues with overall data quality and a lack of standardization” that make data unreliable, the federal procurement ombudsman warned in July.