Working With Code in 2026

At the time of writing, there are many ways to work with code and agentic AI. Here, I am talking only about IDEs, setting aside deployment infrastructure.

I've had good luck with Zed, which is written in Rust rather than being a fork of VS Code.

I've also worked a lot with the macOS Claude desktop app, which hides most code details but has a useful interface for interviews.

One advantage of using Zed, or another third-party system, is the ability to use multiple foundation model and LLM providers.

Using multiple providers reduces vulnerability to acute service outages and makes it possible to orchestrate multi-model workflows.

For example, if Codex models have a benchmark advantage, one workflow can invoke Claude and Codex agents to work in independent worktrees or conduct separate reviews.

This allows convergent solutions to emerge more readily from the aggregated model weights and harnesses.