From Copilots to AI Coding Agents
AI coding agents are autonomous coding tools that combine code generation, context awareness, and task execution, moving beyond simple autocomplete suggestions to coordinate multi-step engineering work across files, tools, and environments. The arrival of Kai for Delphi and C++ Builder and the Devin Desktop platform underlines how agentic AI is evolving along two tracks: platform-specific integration and team-centric orchestration. On one side, Embarcadero is bringing agent-based assistance straight into RAD Studio to support Delphi C++ development, an area often overlooked by mainstream AI tools. On the other, Cognition’s Devin Desktop focuses on engineering workflow automation by coordinating multiple agents across projects and codebases. Together, they show how AI coding agents are becoming part of day-to-day development, whether a team relies on long-lived legacy systems or modern, cloud-based collaboration.
Kai: Agentic AI for Delphi and C++ Builder
Kai is Embarcadero’s agentic AI assistant for RAD Studio, delivered as an extension so that Delphi and C++ Builder users can add AI capabilities to an existing IDE. It depends on external LLMs, either cloud-hosted or local, with developers supplying their own API keys; despite that, Kai is sold as a subscription product at USD 249 (approx. RM1,160) per developer per year. Once enabled, it offers chat, AI-driven code completion, and an MCP server so other AI agents can communicate with the IDE. That matters in a niche where Delphi is still used by 2.5 percent of developers in surveys but powers long-lived, performance-sensitive systems. According to Embarcadero’s Stephen Ball, customers value “fully compiled native code” for tasks such as stock exchange and high frequency trading, and Kai aims to bring AI coding agents into that environment without abandoning legacy compatibility.
Devin Desktop: Coordinating AI Agents Across Teams
Devin Desktop is Cognition’s effort to unify AI coding agents across engineering workflows, combining a full editor with an agent management dashboard. It extends the earlier Windsurf environment so the same Devin agent can run on desktop, in the cloud, and via the command line, giving teams a consistent interface for autonomous coding tools. A central feature is Spaces, which groups agents, tasks, pull requests, and files by project so context can be shared instead of reset each session. Devin Desktop also supports the Agent Client Protocol (ACP), letting third-party or in-house agents run alongside Devin inside the same workspace. According to Cognition’s Theodor Marcu, engineering leaders now must “manage a growing fleet of agents,” and Devin Desktop answers this by providing a single place to coordinate, monitor, and direct multiple AI coding agents within complex engineering organisations.
Platform-Specific vs Team-Centric AI Workflows
Kai and Devin Desktop map to two distinct strategies for AI coding agents. Kai focuses on platform-specific integration: it embeds AI into RAD Studio so Delphi C++ development teams can chat with an agent, generate code, resolve build errors, and even run file operations within the IDE they already use. Its MCP server hints at future interoperability, but the core experience remains centred on individual developers inside one tool. Devin Desktop, by contrast, is team-centric. It treats agents as collaborative participants in a broader engineering workflow, linking multiple ACP-compatible agents and planning an agent router to direct tasks based on cost and performance trade-offs. For teams, the choice now depends on stack and structure: deeply invested Delphi shops gain AI inside their legacy tools, while cross-functional teams gain a hub to orchestrate diverse autonomous coding tools at scale.
