The interview was conducted by Ula Rutkowska (Senior Researcher, Apolitical) and edited by Christina Obolenskaya (MSc in International History, LSE and Communications Intern, Apolitical). Steve Rennie was named on Apolitical’s Government AI 100 2025.

Steve Rennie’s team faced a problem: How do you help farmers in Canada find information on agricultural services and funding on an unwieldy government website? The answer? A generative AI chatbot.

As Director of Data-driven Technologies, Agriculture and Agri-Food Canada, Rennie assembled a team to create a generative AI chatbot for farmers and agricultural workers, winning the Public Service Data Challenge. Read this article to learn how by starting small you can help scale AI innovation across government programmes (and maybe win awards in the process).

Your team won the coveted Public Service Data Challenge award for your AI chatbot – how did you come up with the tool? What were some of the challenges you faced in implementing the AI tool on a community level?

It's a generative AI chatbot that sits on top of our website called AgPal. It's a way for farmers and others in the agricultural sector to find information on government programmes, funding, and services. It was an excellent resource for its time, but it could sometimes be challenging to find information on it. We decided to use generative AI because AgPal already had a well-curated and maintained data set with strong metadata.

If I'm a farmer in Prince Edward Island and I grow potatoes, it would tell me what funding programmes are available. The driving factor was that people have limited time and attention. If you're working in a field for 14 or 16 hours, the last thing you want to do is spend hours on a government website looking for information. It would be much better if you could just ask questions conversationally and get the information you need. Another benefit is that you can get the information in the way you prefer, such as in bullet points or at a different reading level.

The project started with a competition in Canada called the Public Service Data Challenge, modelled after a similar one in the UK. The Data Challenge came to Canada for the first time at the end of 2022, and my department got involved. I was part of a team primarily made up of employees from Agriculture and Agri-Food Canada, but also from other departments across the government.

The idea was conceived by someone in our policy branch. We had people from programmes, technology, and various other areas. It was a great opportunity to bring together public servants to solve problems. One of the first challenges we identified was that getting information to people quickly was important. Around this time, generative AI was becoming mainstream with ChatGPT being released in the fall of 2022. We thought it was an interesting technology and a good use case for our site. We didn't have all the expertise in-house, so we brought in students studying AI from a local college in Ottawa, as well as industry experts. We held a hackathon to unpack what this could look like.

We won the Public Service Data Challenge in the spring of 2023 and spent the next few months getting AgPal Chat ready to launch. We did that in the spring of 2024. Being the first department to launch a generative AI tool to the public, there was uncertainty about the path to production. We had to figure out the process as we went, which was challenging. We needed to determine who to speak with, what approvals to seek, and who to engage with. We also imposed a higher degree of scrutiny on ourselves to ensure we had a useful, accurate, bias-free, accessible, and bilingual product. We did extensive red teaming, user testing, feedback sessions, and testing for accuracy, accessibility, and bias mitigation.

We worked collaboratively across our department and the Government of Canada. This included working with privacy, accessibility, communications, legal services, cybersecurity, and other specialists. We also worked with the Treasury Board of Canada Secretariat and the Chief Data Officer of Canada to ensure we had all perspectives. The initial feedback has been very positive, with high user satisfaction ratings and anecdotal feedback indicating that people find it easier to get relevant information quickly. It was also great to see AgPal Chat mentioned last year in both the Clerk of the Privy Council’s report to the Prime Minister and the Government’s Fall Economic Statement as an example of how AI is being used in the public service to deliver results for Canadians.

Many public servants are hesitant to start using AI. How did your team manoeuvre the early uncertainty of figuring out the safeguards and permissions for using generative AI?

We started with problem definition, focusing on making information more accessible and easier to find. Initially, we considered a glorified search bar, but it morphed into generative AI as we explored options. At the time, generative AI wasn't widely understood in a government context. We framed it as a proof of concept and part of the competition. We talked about it a lot, explaining what we were doing and why, seeking input and buy-in from senior management and stakeholders. This communication was critical to the project's success. We also had supportive senior management interested in growing our AI practice.

How was cross-sector collaboration important for this project?

I have a small team with limited capacity and resources. Just down the street, we have students at the cutting edge of technology who are keen and excited to get involved. It was a no-brainer to work with them. We contacted the college's programme coordinator and invited them to be part of the hackathon. This collaboration snowballed into larger projects. We partnered with colleges on non-critical projects that made great use cases for class projects. The colleges were eager for real-world examples, and we retained data ownership and IP. This collaboration yielded great projects and allowed us to spot talented students for internships and potential hires, creating a talent pathway.

What are some future applications of AI in the agricultural sector?

Agriculture is a fantastic playground for AI given how much data are available. We can analyse data on soil, weather, and crop health to optimise energy consumption in greenhouses. AI can analyse historical weather patterns to predict crop yields, potential disease outbreaks, and weather events. Smart tractors can use AI to optimise the amount of grain or fertiliser applied to specific areas, targeting where it will have the best chance of growing.

What guidance would you offer other government agencies or organisations looking to develop and implement AI-driven solutions?

Start small and work on a contained project that you can deliver. There's sometimes a tendency to think of all the things you could potentially do with artificial intelligence, which can be daunting and ultimately may not be possible. You'll have much greater success if you focus on specific use cases that can show value and then build out incrementally from there. You need to socialise your work. Communicate what you're doing, why you're doing it, and how it will help people.

If you think you've spoken to enough people, you probably only scratched the surface. You need to do at least ten times the amount of talking that you think you've done on this work. Keep the messages simple and clear, stressing the benefit of why you're doing it, not the technology itself. Repeat that message to as many people as possible.

Learn and borrow from others who are already doing this work. At Agriculture and Agri-Food Canada, our approach to AI has been to share our tools freely and widely across the public service. It's not helpful if every department is making a similar version of the same thing. It's not a good use of time or resources. If we've built a chatbot and gone through security assessments, developed technical documentation, and created communications material, why should every other department have to do the same for their chatbots? Sharing these resources frees up other departments to focus on different things they can share across the public service, helping us grow our AI capacity as a government.

What advice would you give to a team leader trying to build a culture within their team? How do you build an AI-ready culture in that way?

It's about finding where people are with their level of AI comfort and allowing them to experiment and test the technology. I'm lucky to have a team keen on the technology, excited about it, and looking for ways to use it. For other teams, it's about helping people realise the value and benefit they'll get from it and making it easy for them to use AI technology. Reduce the friction to using AI tools and integrate them into workflows.

Don't frame AI as a magic wand. There's a perception that AI can solve every problem, but that's not the case. It's one tool among many. Sometimes AI is the best solution, but other times, it's a different technology or not. It could be changing processes or the way things are done. Help people understand that AI is a tool to use as part of their work, just like a spreadsheet or word processing tool. This approach will lead to more success.