AI coding assistants just leveled up, again…

AI coding assistants are rapidly becoming more integrated, more context-aware, and more workflow-friendly—especially inside IDEs and cloud environments—making “AI as a coding copilot” feel less like a novelty and more like day-to-day infrastructure. The most consequential shift is that multiple major tools have landed meaningful upgrades in a short window, pushing beyond inline autocomplete toward chat, project-wide context, and tighter ties to real development resources.

Google’s Duet AI is now generally available and can be installed directly into an IDE. It offers both a dedicated chat panel and inline suggestions/IntelliSense-style completions, aiming to make coding conversations and code generation part of the normal editing flow. Its standout feature is cloud integration: developers can link a Google Cloud project from within the IDE, manage cloud resources there, pull documentation, and generate code samples tailored to that project. Google is also building a cloud IDE called Project IDX (based on VS Code) with Android and iOS emulators running on cloud VMs, reducing the local setup burden that often slows experimentation.

JetBrains has also launched an AI assistant positioned as a “Copilot killer,” though early reviews are mixed. The key differentiator is the JetBrains AI Service, which can power the assistant with multiple models rather than relying on a single underlying system. That opens the door to using a company’s own fine-tuned model—potentially based on open-source code models like Llama or other large code-focused systems—so responses can better match internal coding conventions and expectations.

Meanwhile, GitHub Copilot has leveled up: it now uses GPT-4 instead of GPT-3.5, adds a dedicated chat window, and introduces a “workspace” command that searches across all code files in a developer’s workspace to generate answers with proper project context. The practical payoff is faster onboarding into unfamiliar codebases and better explanations of “spaghetti code.” Copilot can also write commit messages and is available in the terminal for explaining and refactoring commands.

A recurring risk with AI coding tools is generating code from sources developers can’t legally use. Copilot’s new approach helps mitigate this by searching across billions of GitHub files and returning relevant licenses for similar code, improving confidence that suggested snippets are usable.

Despite the momentum, these tools still don’t fully replace developers: they don’t reliably run code or create entire new projects from scratch. The next step, hinted at by current capabilities, is end-to-end execution—turning requirements into runnable Django apps, generating files, running unit tests, and iterating through prompts until the result matches the spec. Image-to-code is another missing piece, with screenshot-to-code tools likely to become IDE-native.

Even if programming becomes less central over time, the demand for engineers doesn’t disappear. The transcript frames coding as a means to an end: future work still needs problem-solvers for robotics, chips, quantum computing, and other frontier technologies. The near-term takeaway is clear—AI assistants are getting more capable and more embedded, and that changes how quickly developers can understand, modify, and ship software.