Accelerating idea to delivery

Over the past seven months, my team and I have focused on accelerating the journey from idea to delivery using AI in software development. In this article, I'll concentrate on the critical first phase: understanding what needs to be delivered at both a business outcome and technical detail level.

This focus has the potential to significantly boost agile ways of working, potentially rendering the "thin slice" concept—designed to limit the need for upfront detailed information—redundant, as information from a single meeting can be transformed and validated into rich, detailed inputs.

Historically, we've grappled with translating high-level business requirements into technically meaningful inputs for development teams. Addressing all potential gaps in this process has been a persistent hurdle, causing delays and unnecessary expense.

We tackled this challenge by focusing on two common problem statements:

The need for Innovation: "We need a new solution to support a change in our business process or to enhance a function we'd like to offer our customers. We've got a rough idea about how it needs to work, and minimal understanding about the current system."

Modernising and managing legacy systems: "We have a legacy piece of code—we're unsure of its internal workings, but when it fails, it becomes mission-critical and can cripple our organisation. It's like an anchor preventing change. We need it to operate reliably and adapt to our changing needs."

Our goal was clear: move from idea to delivery quickly and efficiently, ensuring outcomes are standardised, easy to support, and resilient for operational stability and predictability. It's also crucial for business users to understand what's needed and know what to expect.

Traditionally, addressing these needs required extensive discovery, especially of pre-existing processes. Subject matter experts spent significant time drafting highly specific implementation guides, either to envision the future state or confirm past actions. For example, consider a custom application used daily to manage client information, process work, update systems for payments and records, and notify third parties of next steps. The problem arises when no one fully understands how it currently operates or needs to operate in the future, a situation that can result in a couple of month discovery spiralling into a multi-year effort.

Granular details also need to be specified before development, such as field validations:

• How many characters are allowed in a name?

• What special characters are permitted?

• What are the data types and rules for combining or separating data from two sources?

• When errors occur, how are they managed?

• Does the system retry automatically?

• Does it apply corrections or escalate to human intervention?

• What security policies protect the organisation from malicious actors?
  

These aspects may not have been considered or documented before because the original experts were available, and for business users, the system "just works." However, they're crucial details that cannot be overlooked to meet user and developer expectations and ensure everything continues to function as expected. Or just to answer the granular questions that an analyst who doesn’t read code might not be able to find out without significant engineer support.

This is where upfront discovery processes supported by AI become critical for acceleration and meeting the objective of "sooner, safer, better." In an environment where efficiency and effectiveness are being measured by return on investment, there needed to be a better way.

Transforming detailed, perspective-specific knowledge into assets that meet business strategic outcomes, technical design requirements, and development and testing details is essential. Often, essential information exists but is buried within mountains of code or documentation. Manually extracting these technical requirements is inefficient and depends on individual skills or awareness of organisational policies—a challenge exacerbated by temporary workforces like contractors.

Multiple roles such as designers, analysts, implementation teams, and managers then attempt to interpret the "source of truth," each from their own perspective, leading to diverse interpretations of what's required. This is where AI accelerators can understand and extract the relevant details for validation by subject matter experts, making the process exponentially more efficient and saving up to 80% of people's time.

People bring invaluable, context-specific insights to these accelerators, taking the synthesised information and validating it against modern design patterns and future business processes. AI then supports this process, tailoring outputs to ensure consistent yet specific results for different purposes.

Using AI in this way has also been empowering from a leadership standpoint and offers insights into the future direction of software development. Comprehension of code will no longer be an esoteric skill understood only by a few. Those without multiple programming languages under their belt can interact more meaningfully, accessing what's been done and exploring how we could change, update, and rethink our business processes for the better.

We're now able to create accessible pathways to technical information in natural language. This means you can ask questions and get answers in a way that makes sense to you, from a reliable source of truth. This is an exciting next step in accelerating growth and learning by applying AI to the SDLC. It has the potential to reduce work that used to require a large team down to a rapid cycle of discovery, reuse, adaptation, and streamlined understanding of what's needed for new development to meet customer needs, and critically in this current fiscally constrained environment, faster time to value at reduced effort.

What's next?

In this series of articles, I'll share the learnings and insights we've gained throughout this journey. We'll explore how our AI-enabled approach is transforming our ways of working and what this could mean for the wider software development industry. I'll look at the thinking we’ve been doing on AI scaling at an organisational level, discuss common pitfalls we've encountered, and offer guidance on how teams can overcome them.