IDE vNext: Who Will Build a Home for AI Agents? (Series #AI4SDLC)

At DotNext I also took part in a roundtable about IDE vNext. The recording, discussion summary, and a separate longread with the position I prepared for the panel are now available. You can find them all on the materials page on my site. The discussion took place on September 25. Dmitry Ivanov, Andrey Berezniker, Oleg Chirukhin, and I talked about what a development environment should become when agents can take on part of the implementation. Anatoly Kulakov moderated.

The question quickly goes beyond an editor with a chat window. An agent can write code faster than a person can tell whether it solved the task correctly. What should the IDE help us do then?

We discussed a few things:

🔸 Which parts of engineering work to start from Pick a task, describe its constraints, delegate it, verify the result, and understand what happens once it is running.

🔸 Where familiar tools work better A keyboard shortcut for finding a symbol’s usages can be faster than explaining the request to a model. Precise renaming and probabilistic generation also work differently.

🔸 How to keep track of several agents More terminals don’t give a person more attention. You need to know which change needs a look and when to step in.

🔸 What to do about shared specifications and platforms Independent changes can often be merged, while incompatible architectural intentions require people to reach an agreement. Agents also need context about services, dependencies, and environments.

We didn’t draw a single picture of the ideal IDE. We disagreed about the role of code and the ways people should interact with it, which makes the arguments worth comparing. What do you most miss in your environment for working with agents today, and what task does that make harder?

#AI4SDLC #AI #IDE #Engineering #PlatformEngineering