As promised in my “Happy New Year” post, I wrote an article on adesso’s blog that shares how we supercharge enterprise applications with AI: https://www.adesso.de/en/news/blog/code-execution-oder-domain-specific-language-dsl.jsp
We argue that the current wave of agentic AI will only pay off in enterprises if it delivers real efficiency gains without sacrificing security, compliance, and control.
Building on Anthropic’s “Code Mode” idea—compressing many tool calls into on-the-fly, AI-generated code—we propose a safer alternative: “DSL Mode”. The model outputs a workflow in a constrained DSL, and a deterministic interpreter validates and executes it, keeping the execution surface small and auditable.
From there, we outline tool discovery to cut context further and, in some scenarios, use GraphQL as a powerful, well-defined interface that can be kept secure—culminating in ten concrete principles for enterprise-ready agentic systems you can actually run in production.
Exciting 🤓;
This article has been written with help from an LLM, which may make errors (like humans do 😇).