Session notes
AI for Leaders: key points from the session
Summary of the AI Mavericks online session AI for Leaders, held on 12 August 2026 with Raghid W. Charara, Senior Executive, AI & Data at Qatar Development Bank. These are notes from the recording, not a verbatim transcript. Each point links to the moment in the video.
Published 4 October 2026
Quick facts
- Session
- AI for Leaders, AI Mavericks, online, 1 hour
- Date
- 12 August 2026, 5:00 PM Qatar time
- Speaker
- Raghid W. Charara, Senior Executive, AI & Data, Qatar Development Bank
- Recording
- AI for Leaders on YouTube
Key points
- What AI is made of (3:24): Raghid W. Charara described AI as hardware (the GPU), software (the large language model) and data. He stressed that data must be clean, stable and ethical: "garbage in, garbage out", however strong the model is.
- AI can be biased, so the user is the filter (8:50): Using a Picasso image-training example, he showed that a model can be wrong because of how it was trained. His rule of thumb: never trust AI blindly, verify before you copy and paste, and use other models to check an answer.
- How AI reads a sentence (9:56): AI breaks text into tokens and gives them numerical weights and vectors, which is how it picks up intent even when people write in different styles or make typing mistakes.
- Large, small and vision language models (14:14): He described large language models as "Mr. Know it all", small language models as subject-matter experts trained on one narrow area such as credit cards in banking, and vision language models as the ones that read images and documents.
- Human in the loop (17:29): He explained human-in-the-loop with the banking maker-checker idea: the AI prepares and suggests, the person approves, rejects or returns. Once an agent is tested and trusted, it can run more autonomously.
- Guardrails (21:43): Guardrails are instructions that keep an AI agent on its task. His example: a travel-agency chatbot without guardrails will still answer an unrelated question such as the purpose of life.
- Prompting and audience (24:51): He discussed the saying that AI will not replace you, but a person who knows AI might, and showed how to rewrite rough notes for a given audience and tone.
- Define the problem first (36:36): Quoting the idea that with an hour to solve a problem you should spend 55 minutes thinking about the problem, he urged leaders to understand the problem and its impact before choosing a tool.
- Building an agent in plain English (39:55): Live in Copilot, he built a market-analysis agent by writing instructions in normal English, with no coding, then refined the instructions and asked the agent to show its sources (see 45:22).
Watch from the key moments
- What AI is made of - from 3:24
- AI can be biased, so the user is the filter - from 8:50
- How AI reads a sentence - from 9:56
- Large, small and vision language models - from 14:14
- Human in the loop - from 17:29
- Guardrails - from 21:43
- Prompting and audience - from 24:51
- Define the problem first - from 36:36
- Building an agent in plain English - from 39:55
More
Questions about AI Mavericks sessions: community@aimavericks.live