What AI Coding Agents in the IDE Really Mean
AI coding agents are software components that use large language models to monitor, understand, and modify code directly inside your IDE, helping with tasks like editing, debugging, refactoring, and project automation without leaving the development environment. The key shift is that these agents are no longer separate chatbots running in a browser tab; they are being wired into the same debugger, profiler, and test runners developers already rely on. This kind of IDE integration changes AI from a one-off assistant you query when stuck into a constant presence that can react to code changes, build failures, or performance issues in real time. For developers, the big question is whether this native presence accelerates work or adds noise, and whether toolchains become simpler or more fragmented as new AI layers arrive.
Visual Studio AI: Agents in the Debugger and Profiler
Microsoft is turning Visual Studio AI from a side panel into a first-class part of the core toolchain. At Build, the company described agents that participate in development rather than sit next to it, embedding GitHub Copilot-based helpers into debugging, profiling, testing, merge conflict resolution, and .NET modernization workflows. Instead of pasting stack traces into a chat window, an agent can inspect live state in the Visual Studio debugger, explain why a service is slow under load, suggest potential fixes, and help validate them in context. Model flexibility is a second major change: Visual Studio is moving toward a bring-your-own-key model, where teams can connect different AI models, running locally or in the cloud, rather than being locked to a small set of sanctioned endpoints. According to Mads Kristensen, this is meant to support teams whose environments made previous AI integrations a non-starter.
Kai Brings Agentic AI to RAD Studio’s Niche
While Microsoft bakes AI into Visual Studio, Embarcadero is taking an extension-first route with Kai for RAD Studio, which serves Delphi and C++ Builder developers. RAD Studio itself does not ship with AI; installing the Kai extension unlocks chat, AI-powered code completion, and a Model Context Protocol (MCP) server that lets external AI coding agents communicate with the IDE. Kai relies on third-party large language models, running in the cloud or locally, and developers must supply their own API keys. It is a subscription product costing USD 249 (approx. RM1,165) per developer per year, with free trials available. The agent can generate code, resolve build errors, manage version control, perform file operations, and talk to other MCP servers from inside the IDE. That makes Kai a relatively minimalist but meaningful step toward IDE integration in a tool that still appeals to teams focused on compiled native code and performance-sensitive applications.
Wearable Terminals: Smart Glasses as AI Coding Workstations
AI coding agents are also driving new hardware form factors. Monako Glass packs a Linux-based system, waveguide display, camera, speakers, and AI coding-agent connections into a 48-gram pair of smart glasses. The company positions the glasses as a wearable command layer for technical and creative tools, naming Claude Code, Codex, Unreal Engine, Blender, and After Effects among supported workflows. Rather than trying to replace a laptop, Monako’s more plausible use case is as a wearable terminal: a developer can check agent-driven progress, approve steps, send prompts, or review outputs without returning to a full desk. A bone conduction microphone captures nasal vibrations to help in noisy environments, while the Vision Engine turns small gestures into commands. The concept hinges on basics like battery life, display readability, and input accuracy, but it shows how developer tools AI may extend beyond the traditional monitor-and-keyboard setup.
From Separate Tool to Native Layer in Developer Tools AI
Taken together, these moves show AI coding agents becoming a native layer across developer tools, not an optional add-on. Visual Studio’s agents live inside the debugger, profiler, and modernization pipelines; Kai gives RAD Studio users an IDE-aware assistant with MCP connectivity; Monako Glass experiments with a wearable terminal that keeps agents within glance and gesture distance. For productivity, the upside is fewer context switches: AI can see what the IDE sees, from call stacks to project structures, and respond in place. For tool consolidation, the trend cuts both ways. Some workflows may centralize around a single IDE integration, while others may add new layers—extensions, MCP servers, wearable terminals—that need configuration and governance. Teams will need to decide where AI belongs in their stack, which models are acceptable, and how deeply they are comfortable letting agents act inside core development workflows.
