Agent based tools like Cursor and Lovable are reshaping how developers approach software. But for data and AI development, the new vibe coding tools fall short. That was the focus of my talk at the Data Science Festival in London earlier this month.

Those tools work well for traditional software problems with clear inputs and clear outputs. Data work does not follow that structure. It is exploratory. You shape questions, work with shifting schemas, and adapt to evolving goals. That complexity requires something different.

At Zerve, we have built for this reality.

Data work is not just code

In most software environments, agents operate within predictable patterns and tightly scoped tasks. With data projects, the path is less predictable. You work with changing schemas, shifting goals, and outputs that evolve over time. That makes standard code agents less helpful for data and AI development.

To work effectively, agents need access to more than the code. They need to understand your data, your environment, and your workflow. That is where Zerve comes in.

Context aware agents are essential

Zerve provides a development environment where agents operate with full context. They can see your files, your data sources, and your code blocks. This allows for better planning, execution, and iteration.

I demonstrated how planning agents create milestones and break them into tickets. Each ticket is assigned to a code agent that runs independently. Everything is cloud based, and compute is allocated automatically. This enables parallel workflows without extra setup.

Agents are not replacing data scientists

This is important. Agents are not here to replace experts. They are here to help them work faster. People still provide direction, domain knowledge, and quality control. Agents write code, evaluate patterns, and handle repetitive tasks. They need human input to stay on track.

I also shared how my mom, with no coding background, used Zerve to plan a vacation by prompting the agent and generating a travel plan. It worked.

Reflections from the event

It was great to see how many young people are already working on real problems in data and AI. The energy in the room stood out. I also had a few conversations about creative tooling. Several folks mentioned using CRIU for Docker, which I found interesting.

I do not love watching myself on video, but in case you missed it, the full talk is linked below.

Let me know what stood out to you. And of course, try our Community tier.