Code in the Age of Agents: Bottlenecks, Languages, and the New Software Reality

Speaker: Dan R Ghica (Programming Langauge Laboratory, Boole Research Centre, Huawei Research UK. University of Birmingham, School of Computer Science)
Series: Logic and Semantics Seminar (Computer Laboratory)
Note: Note unusual time

Everything has changed, and the software engineering landscape will never look the same. Autonomous coding agents can already produce vast implementations, dismantling the long-held assumption that writing code is the primary constraint of software development. As the marginal cost of code generation becomes negligible, where does the real bottleneck relocate? Is the new point of friction in comprehension, review, and verification, or does the critical challenge move even higher up the stack, in specification and validation?

Furthermore, agent-generated code differs fundamentally from human-written systems in style, cognitive idioms, and architectural failure modes (e.g. slop: plausible nonsense, context drift, abstraction leaks). This shift forces a fundamental reconsideration of our conceptual toolkit: What constitutes a suitable programming language for an agentic era? Which language features become essential to address these emerging bottlenecks and failure modes, and which legacy ergonomics designed for human cognitive limits are rendered largely irrelevant?

The talk, which will barely scratch the surface of such grand and rapidly evolving issues, will be informed by a dual perspective: industrial programming language design and development, and the academic research of the theoretical aspects of this topic.