AI‑Driven Software Engineering: From Requirements to Operations in the Era of Software Engineering 3.0

2023 is marked as the inaugural year of Software Engineering 3.0, introducing a new paradigm called ML‑DevOps (or more precisely LLM‑DevOps), where large models drive both development and operations, extending the traditional software lifecycle with AI capabilities.

The piece traces AI’s lineage from Turing’s 1950 paper and the 1956 Dartmouth conference, through the evolution of GPT models (GPT‑1 to GPT‑4) and early AI applications in software testing dating back to the 1970s, highlighting a gradual but steady integration of AI into software engineering.

In the requirements phase, LLMs enable automatic generation of SysML models from natural‑language specifications, and research directions include Chinese‑language requirement modeling, ML‑system requirement decision making, user‑feedback mining, and traceability recovery techniques such as GeT2Trace.

For design, AI assists in design generation (e.g., Altair DesignAI), design exploration (e.g., Design Explorer), and optimization (e.g., Synopsys DSO.ai), allowing engineers to explore high‑performance alternatives, automate repetitive tasks, and reduce computational costs.

Programming is enhanced by intelligent code completion, which builds language models from source code corpora and uses contextual similarity to suggest completions; approaches range from identifier‑sequence and abstract‑syntax‑tree representations to statistical N‑gram and neural models, with tools like aiXCoder illustrating practical adoption.

Testing benefits from AI‑driven techniques that improve test‑path generation, data diagnostics, and assertion accuracy, especially in GUI testing via computer‑vision and OCR; notable platforms include Test.AI, Applitool, Mabl, AirTest, and others.

Operations are empowered by AIOps, which provides three core capabilities: comprehensive real‑time monitoring, knowledge extraction from massive ops data to drive automated actions, and fully automated operational workflows.

The article concludes that AI has already played a significant role across software engineering 1.0 and 2.0, and with the release of GPT‑4 it promises a future where much of the design and coding can be automated, while also promoting a related Python‑based continuous deployment course.