Title

Agile in AI or AI in Agile?

Overview

Developing AI solutions requires a detailed examination of available data, thorough analysis of solution alternatives, and repeated hypothesis testing to determine the best approach. The heavy data intensive exercise to build appropriate data models under high degree of uncertainty, ambiguity and complexity requires unique non-traditional development practices. While Agile methods in general are appropriate in complex environments, not all of the Agile software development methods and techniques work for the AI domain. There are various custom tailored Agile techniques and practices which could be applied while developing AI solutions like DoR for ensuring sanity of data, Lean Startup based Agile approach is customized to validate evolving data models early in the cycle to avoid waste of time and efforts. ModelOps is picking up a systematic way to quickly and responsibly develop, test, and deploy AI models. The evolving AI lifecycle development methodology does a fine balancing act between application-focused agile development and data-focused data methodologies as both are required for AI solution development. There are some illustrative Methodologies practiced by various innovative companies which have evolved from CRISP-DM established originally by IBM.

The other side of the story is how and where AI and ML techniques particularly are helping businesses in better Agile and DevOps adoption. There are cases where NLP is being used to establish initial Product Backlog as it can effectively parse various sources of unstructured data and create initial Product Backlog for Product Owners. AI techniques are coming very handy in implementation of Continuous Testing where dependency on testers is being reduced significantly. Telemetry and Continuous Monitoring is getting benefited by adoption of sophisticated artificial intelligence (AI) models working in the extraction of insights from vast data sets. These models can help detect unusual behavior that could lead to security flaws and failures. IT Ops, DevOps, and SRE teams are able to work smarter and faster using increasing adoption of AIOps where ML and NLP techniques are used to detect digital-service issues earlier and resolve them quickly, before business operations and customers are impacted.