Last night, we were thrilled to welcome over 150 attendees to our offices for the 12th Data & AI London Meetup. This session featured OpenAI’s Katia Guzman, who shared insights on building agent-based systems using the latest ChatGPT models.

‍

Katia, a Developer Experience Engineer at OpenAI, brings extensive experience in AI and developer tools, having previously worked as a Solutions Architect at OpenAI, a startup CTO, and an engineer at Stripe. She walked us through the power of AI reasoning, demonstrated deep research capabilities, and answered key questions on extracting value from ChatGPT.

‍

‍

Key Takeaways from the Event

‍

1. Advancing AI Reasoning: Thinking Before Speaking

‍

Modern AI reasoning models now "think before they speak" – engaging in internal deliberation before generating responses. Instead of just predicting the next word, they evaluate options, refine outputs, and structure responses with greater intentionality. This inspectable thought process builds trust, transparency, and reliability in AI-generated insights.

‍

2. AI Excels at Navigating Complex Tasks

‍

AI is better than ever at handling large-scale, intricate challenges, including:

• Finding a Needle in a Haystack – Isolating crucial insights from massive datasets.
• Analyzing Complex Information – Processing vast amounts of unstructured data.
• Multi-Step Planning – Moving beyond simple Q&A to autonomous, goal-oriented execution.

‍

This shift positions AI not just as a knowledge provider but as an independent problem-solver.

‍

‍

3. The Evolution of AI Agents: From Answering Questions to Taking Action

‍

AI is transitioning from passive responses to active execution. OpenAI’s Operator and Deep Research are leading this shift:

• Operator (coming soon via API) automates actions, reducing manual intervention.
• Deep Research generates detailed, source-cited reports, ensuring credibility and transparency.

‍

These advancements mark a new era where AI doesn't just provide answers—it gets things done.

‍

4. Structured Data & Intelligent Agents: Unlocking New Possibilities

‍

At the event, we saw AI transform unstructured data into actionable insights in three steps:

1. Leverage Inaccessible Data – AI makes sense of raw, unstructured information.
2. Turn Data into Structured Insights – Creating usable formats like graphs and tables.
3. Use Reasoning to Build Smarter Agents – Enabling better decision-making and automation.

‍

With GPT-4o mini, AI can autonomously determine the best way to build applications, reducing the need for manual development.

‍

5. Ensuring AI Reliability: Validation & Continuous Testing

‍

For AI to be effective in production, validation is key. Reasoning models help by:

• Enhancing Entity Validation – Detecting inconsistencies and providing context.
• Continuous Testing with Evaluation Sets – Ensuring models remain stable as they evolve.

‍

Rather than a one-time fix, robust evaluation frameworks help AI systems adapt and improve sustainably.

‍

Join the Data & AI London Community

‍

We just hit 3,500 members in our Meetup community! I started this group to provide engineers and data scientists with a space to share knowledge, network, and learn from industry leaders in a relaxed setting.

‍

Over the past three years, we've hosted experts from OpenAI, Nvidia, Google, Databricks, Peloton, GoCardless, EDF, and Mesh-AI. Looking ahead to 2025, we have at least five more meetups planned—plus some incredible speakers!

‍

📅 Next up: Women in Data & AI London – March 26
‍

🔗 Sign up for our Meetup group here to stay updated on future events.

‍

Let’s keep pushing the boundaries of AI innovation – together. 

‍