Are you excited about AI - or feeling overwhelmed? Do you see it as a chance to achieve more for your organisation; or are you put off by technical barriers, ethical issues, and stories about slop, bias, and backlash?
Inspired by our 2025 series introducing the risks and opportunities of generative AI for organisations like charities, CICs, and not-for-profits, we’ve added information about agentic AI, coding agents, and smaller models.
Our webinar covers what agentic AI coding agents are, the difference between AI, genAI, machine learning, and data science, and what charities need to know about these tools. We discuss how GPTs actually work, how that influences their capabilities and limitations, and how they use your data, with practical use cases for genAI in the poverty and social justice sectors, and ethical considerations.
Q and A
How do programmes like Magic Notes for Case Notes balance client rights and data protection?
The ethical aspect requires careful consideration. Some of these tools promise no data will ever leave your account, but it is still a risk. You can also ask for any data to be deleted after you have used the notes. Use popular tools from well-known companies to reduce the risk of data breaches, and ask for their cybersecurity certifications.
Do AI Agents retain information and memory if you delete the chat where you gave or asked the information?
If you use ChatGPT or Claude, then by default it retains your data, although you can usually opt out in the settings. The question is whether the data is used by them for any other purposes, such as training their models, which you can opt out of if you have an enterprise account. Each tool is different, so look at the T&C’s to be sure.
What security vulnerabilities might occur when coding with AI tools, and what should we look out for?
It depends on what you are using the coding tools for, there’s a different level of scrutiny required when doing data analytics compared to developing software. If you use it for interacting with data, then the risks of data privacy and breach are most severe, and you should try to get a (human) expert’s opinion on how to avoid them.
If you are building software, the best practice is to incorporate cybersecurity tools and best practices from day 0, regardless of whether you use AI for coding. AI coding tools increase your efficiency and speed as a developer, so if anything, the cyber tools and practices are even more important.
Can you create code to create website pages in off-the-peg websites eg Squarespace?
Some of these websites have built in tools for creating websites using generative AI, for example, Wix with Base44.
How does a free tool generate costs, as described in the increased use of ‘tokens’?
The tool itself may be free to use, meaning there is no fixed licence or subscription charge. However, many AI-powered tools charge based on usage, typically measured in tokens (input and output text processed by the model), or sometimes as credits.
The financial consideration arises because the actual token consumption can be difficult to predict. For example, when an AI agent evaluates responses, calls other agents, performs multiple reasoning steps, or invokes sub-agents behind the scenes, additional tokens may be consumed without being immediately visible to the user.
Resources
Your speakers
Natan Mish
Machine learning engineer building production-grade medical AI by day, overthinking Python tooling and error messages by night. At Zimmer Biomet, he helps ship models that support orthopaedic care, but still gets a disproportionate amount of joy from finding bugs before they find users. He has previously wandered through transport, fintech, and even a central bank, collecting data problems and turning them into talks at PyData and EuroPython along the way. Outside of work, he volunteers with DataKind UK — putting data science to work for social good.
Fiona Lau
Fiona Lau is an AI researcher, technologist, and entrepreneur working in the telecommunications sector, where she focuses on research and innovation. She develops proof-of-concepts (PoCs) that apply machine learning, computer vision, large language models (LLMs), and other AI technologies to solve real-world business challenges. With experience across finance, technology, and applied research, Fiona specialises in turning emerging AI capabilities into practical products and solutions. Her work includes AI evaluation frameworks, retrieval-augmented generation (RAG), agentic systems, and data analytics. She is passionate about responsible AI, product innovation, and building technology that delivers meaningful impact.
Dulcie Vousden
Dulcie is DataKind UK’s Head of Data Science, where she has spent the last five years supporting third sector organisations to use their data more effectively and advocating for the ethical and responsible use of data in the sector. Prior to DataKind UK, she completed a PhD in medical biophysics and has over 15 years of experience designing and analysing data-focussed projects.
Thank you to the Insight Infrastructure programme for their support. Developed by the Joseph Rowntree Foundation, it aims to democratise access to high-quality quantitative and qualitative data and evidence through open collaboration and innovation, to help tackle injustice and inequality in the UK.
