Descriptions:
Rajiv Chandegra, a practicing medical doctor in London and AI engineer at Annicha Labs, introduces what he terms “adaptive engineering” — a proposed next paradigm for AI systems that moves beyond the fixed, pre-determined harnesses that underpin tools like Claude Code, Cursor, Codex, LangChain, and Goose. His talk argues that the current approach to AI engineering — designing system prompts, tool definitions, agent roles, and loop structures entirely before runtime — works well for complicated, decomposable problems but becomes brittle when agents encounter dynamic, real-world situations.
Drawing on complexity science and systems thinker Russell Ackoff’s concept of a “mess” (a system of interacting, constantly shifting relationships that cannot be decomposed into independent parts), Chandegra distinguishes between complicated problems (like building a jumbo jet — hard but fully plannable) and complex problems (like a market or human organization — where the whole cannot be derived by analyzing the parts). He argues that treating complex problems as complicated ones is one of the most expensive mistakes in modern engineering.
Adaptive engineering, as he defines it, allows the harness itself to emerge, stabilize, evolve, and eventually dissolve during the runtime of an engineering process. The engineer’s role shifts from pre-specifying every structure to designing constraints and rules of play within which the harness self-organizes. The talk is philosophical and forward-looking, positioning this approach as especially relevant as AI moves from software-only environments into multi-agent, multi-institutional, and physical-world applications.
📺 Source: AI Engineer · Published July 07, 2026
🏷️ Format: Opinion Editorial
