The head of development’s AI decision framework

Artificial intelligence is continuing to change how bespoke software is developed. New tools for all sectors and scenarios promise faster development, improved developer productivity, and increased automation across the software lifecycle.

For Heads of Development, however, the challenge is not understanding that AI matters or that it should be incorporated; it is deciding where to start.

Many organisations face a difficult choice between adopting or testing multiple AI tools quickly or delaying adoption until approaches become clearer or teams are more aligned. In practice, at Propel Tech, we believe the most effective strategy sits between these approaches, and is always a staged and planned approach that starts with the business need and clear goals built on realities, not AI hype.

Successful adoption starts with good bespoke software processes and doing the essential work to understand how development teams currently spend their time and where AI can improve bespoke software development outcomes without disrupting delivery, and while enhancing the role of developers.

Key takeaways

• AI should be viewed as a capability change for development teams, not just a tooling decision.
• The greatest gains typically come from redirecting senior developer time toward architecture and system design.
• AI can significantly reduce junior developer onboarding time, but requires careful sequencing.
• Mid-level developers often benefit most from AI adoption, as it allows them to develop higher-level technical capabilities sooner.
• Organisations should prioritise high-impact, low-disruption use cases when introducing AI into development workflows.

Understanding where development time is spent

Before introducing AI tools as part of the bespoke software process, it is important to understand how development teams currently operate.

We see time and time again that when organisations map development activity across roles and delivery stages, they often find that significant time is spent on work that supports delivery but does not necessarily deliver the highest value.

We are sure you will be aware of these tasks, they are essential but take lots of time and are highly susceptible to human error. These activities include, but are not limited to:

• Manual code review processes
• Repetitive implementation tasks
• Developer onboarding challenges
• Documentation and knowledge sharing

AI can help reduce the time spent on many of these activities which frees up developer time for problem-solving and thinking.

Alongside this function of AI, there is another opportunity in how that time is redistributed across the development lifecycle. At Propel Tech, we have been working to identify the most effective ways to support developers through the AI transition at different stages of the software lifecycle, and we’ve found that there is no one-size-fits-all approach.

Senior to junior developers are being impacted in different ways, and to incorporate AI it's important to understand this, specifically for each role.

The role of senior developers and how AI can support:

Senior developers provide the greatest value when they focus on architecture and system design; they are the key to effectively creating bespoke solutions that deliver to business needs and fixing issues effectively.

Their responsibilities include:

• Defining system structure
• Managing complexity across platforms
• Designing scalable and maintainable solutions
• Guiding technical standards across teams

However, many senior developers spend a large proportion of their time on code review.

While review is essential for quality assurance and mentoring, much of the work involved can be mechanical. Tasks such as validating naming conventions, checking test coverage, or identifying common errors can be increasingly supported by AI-powered tooling.

When these activities are partially automated, senior developers can spend more time focusing on architecture, long-term system design, and technical leadership.

For many organisations, this shift represents one of the most significant benefits of AI adoption.

Improving junior developer onboarding

Onboarding new developers into an existing codebase can be one of the slowest parts of software delivery.

Even experienced developers require time to understand:

• The architecture of the system
• Established design patterns
• Integration points between services
• The reasoning behind previous technical decisions

This learning process can take weeks or months.

AI tools can help shorten this onboarding period by allowing developers to explore codebases more easily, setting out the groundwork, and delivering summaries, ask contextual questions, and generating initial implementation scaffolding for review, adaption and checking.

However, introducing AI too early can create additional complexity. Developers who do not yet understand the system architecture may struggle to evaluate whether AI-generated output aligns with it, and it may cause problems if junior employees act on AI without an expert view.

We find that successful teams typically introduce AI after skilled developers have developed a basic mental model of the system, reviewed the problem they are solving, created a human view and combine this with structured review and mentoring, using AI to support, not to lead, this process.

Supporting mid-level developer progression

Mid-level developers are often highly productive but may spend significant time completing implementation tasks, basic reviews and supportive actions that do not significantly expand their capabilities.

AI can accelerate many of these mid-level tasks and shorten their time spent on basic activities, allowing them to support junior colleges more closely or take on new senior skills quicker.

One core area that AI can be used to support professional development in is by involving mid-level developers earlier in higher-level technical activities, such as:

• Architecture discussions
• Design reviews
• Cross-system decision making

Using AI in this way can help development teams, support midlevel developers by offering a sparing partner and enabling them to focus on high-skilled activities, or providing a second view, allowing them to strengthen their development capability while improving delivery speed.

A framework for evaluating AI opportunities

Alongside starting to use AI and LLM’s it's imperative to have a measured way to track changes.

When considering AI initiatives in bespoke software development teams, it can be helpful to assess potential changes across two dimensions:

Impact – the potential improvement in delivery speed, quality, or team capability.

Disruption – the level of change required to existing workflows.

High impact, low disruption

These initiatives are typically the best starting point.

Examples include:

• AI-assisted code review
• Automated test generation
• Documentation generation
• Developer knowledge assistance for internal systems

These approaches integrate into existing workflows while delivering measurable benefits.

High impact, higher disruption

Some initiatives require broader changes to development practices.

For example, AI-augmented development workflows may involve developers increasingly directing AI-assisted tasks and reviewing generated code.

These approaches can deliver significant improvements but require training and changes to working practices.

Lower impact initiatives

Some AI tools provide incremental improvements such as autocomplete or simple code generation. These can be useful but should not be mistaken for transformational change.

What successful AI adoption looks like

Across organisations implementing AI successfully in bespoke software development, several common practices emerge.

Start with a specific workflow

Rather than attempting to transform an entire development organisation immediately, teams often begin by introducing AI within a specific workflow or project.

This allows experimentation while maintaining control of delivery.

Develop new team capabilities

As AI tools become more widely used, developers increasingly focus on directing AI-assisted work and reviewing generated code. Supporting these capabilities requires structured learning and mentoring.

Measure outcomes

Organisations benefit from tracking key metrics before and after AI adoption. These may include delivery cycle time, review turnaround, or time from development to deployment.

Maintain strong review processes

AI-generated code should be reviewed carefully to ensure it aligns with architecture, security, and quality standards.

Next steps

For organisations exploring how AI can support bespoke software development, the first step is understanding the readiness of their existing systems and processes.

Propel offers a Free AI Software Audit designed to help organisations assess:

• Code quality
• Architecture and data readiness
• Development workflows
• Security and governance considerations
• Opportunities for AI-assisted development

Learn more here:

https://propeltech.co.uk/free-ai-software-audit-assess-code-quality-security-ai-readiness/