Requirements specification has seemed to be forgotten and has lost its significance over the past decade. Agile culture shifted the focus to stories and sprints, and “requirements specification” began to sound like the waterfall model. <\/p>\n\n
The paradox is that writing backlog content is<\/em> specification work; it simply was no longer called by that name. Requirements specification as a distinct skill fell into the shadows, but over the past year, it has been discussed more than ever. In software development, spec-driven development has emerged, where AI-assisted coding relies on carefully written specifications. <\/p>\n\n Requirements specification is on the agenda because, with increased efficiency, the bottleneck in IT projects has shifted from implementation to specification and, on the other hand, to quality assurance. <\/p>\n\n Two different questions are at the forefront of discussions: <\/p>\n\n <\/p>\n\n When defining information system requirements (or writing features and stories, i.e., backlog content), in this blog I am referring to describing new specifications: new functionalities, business rules, or requirements related to new legislation. <\/p>\n\n Defining something new is essentially different from copying an existing system, where the system is renewed by only changing the technical platform. In changing the technical platform, requirements specification work is not done in the true sense of the word; instead, specification is done by copying the requirements of the existing system. <\/p>\n\n AI is a significant help in copying the requirements of an existing system: a foundation for requirements can be obtained relatively effortlessly from the code of an existing system, or the code can be directly translated to a new programming language. However, old code is practically always incomplete and contains errors, so a person must review the foundation produced by AI, supplement it, and correct logical errors. <\/p>\n\n In the specification work of a functionally new system, the possibilities of AI are also diverse: <\/p>\n\n I would like to emphasize that producing requirements still requires discussion between different parties. A cable cannot (at least yet) be pulled directly from one party’s brain to another and allow shared brain chemistry to produce, for example, new business rules. <\/p>\n\n <\/p>\n\n Specification is followed by solution implementation, coding, or configuration. <\/p>\n\n In AI-assisted coding, AI is given a prompt, which can be directly well-prepared specification documentation. The higher the quality of the specifications, the better the AI-generated code. <\/p>\n\n However, universally applicable, precise format guidelines for AI-optimized specification documentation cannot yet be provided. The answer depends on the platform and language model used in coding, the functionalities being implemented, and even the coder’s working methods. It cannot be stated that precisely this structure, format, or Definition of Done always produces code that fully meets the needs. <\/p>\n\n AI-optimized specification sounds like a new question, but the answer is surprisingly familiar: high-quality specification produces good code regardless of whether the programmer is a human or AI. <\/p>\n\n <\/p>\n\n The requirement for high quality in specifications stems from the fact that a human coder knows to ask if something in the requirements remains unclear. AI does not necessarily ask; instead, it guesses. The guess may sound plausible but be wrong. This raises the requirement for unambiguity: the less room for interpretation in the specification, the less AI can fill gaps with its own assumptions. <\/p>\n\n Before reaching well-prepared, prompt-ready specification, a long journey has already been traveled from the customer’s, business’s, and other stakeholders’ needs, problems, and objectives. Along the way, the we have to remember the basics \u2014 the old, almost timeless requirements specification guidelines: <\/p>\n\n <\/p>\n\n Test cases also work well in AI-assisted coding: AI is guided to produce code that passes the given tests. This follows the familiar principle of test-driven development (TDD), which fits particularly well with language models. <\/p>\n\n Good specifications can easily be turned into test cases with the help of AI, and the key here is also, above all, exceptionally high-quality specifications. There is also a risk here: code produced by AI may pass tests but not work correctly outside the test cases. <\/p>\n\n <\/p>\n\n Although AI can offer extensive help with specification, the human role does not disappear. Specifications cannot and should not be created in a fully automated manner. <\/p>\n\n Being a good business analyst requires experience and professional competence. AI can help a less experienced business analyst produce more structured and clearer requirements. However, AI does not replace experience. A beginner easily overlooks, for example, non-functional requirements and other constraints that AI does not know to ask about on its own. <\/p>\n\n Humans, and especially human interaction, are still essential in cases such as the following: <\/p>\n\n <\/p>\n\n Resolving these questions requires human interaction, not AI. <\/p>\n\n AI is an excellent tool in specification work. It makes specification routine work easier, but at the same time, it makes the human role more demanding. The most important thing is investment in specification competence, not AI tools.<\/strong> <\/p>\n","protected":false},"excerpt":{"rendered":" Requirements specification has seemed to be forgotten and has lost its significance over the past decade. Agile culture shifted the focus to stories and sprints, and “requirements specification” began to sound like the waterfall model. The paradox is that writing backlog content is specification work; it simply was no longer called by that name. Requirements […]<\/p>\n","protected":false},"author":18,"featured_media":15435,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[304],"tags":[],"class_list":["post-15436","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-requirements-specification"],"acf":[],"yoast_head":"\n\n
\n
How can AI accelerate requirements specification work?<\/strong> <\/h2>\n\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
How to create requirements specifications in an AI-optimized manner?<\/strong> <\/h2>\n\n
AI-assisted coding requires extremely high quality from specifications <\/strong> <\/h2>\n\n
\n
\n
\n
\n
\n
\n
\n
Human in the driver’s seat<\/strong> <\/h2>\n\n
\n
\n
\n
\n
\n