For consumer and retail technology leaders, the conversation around AI-assisted software development is moving quickly beyond experimentation.
At NashTech’s October consumer and retail working group, technology leaders discussed how AI is already changing the way teams build, review, test and maintain software. The consensus was that increasing development speed creates a new set of questions around what happens next.
David Purves, who leads NashTech's retail and consumer practice, spoke to participating technology leaders ahead of the session. Across those conversations, he saw five challenges repeatedly emerging: AI-assisted development, testing and quality, governance and guardrails, skills and standards, and adoption and knowledge sharing.
What’s becoming clear is that AI-assisted development isn’t an isolated technology change. As development accelerates, technology leaders must look at what happens across the entire delivery lifecycle, from testing and governance to skills, cost and adoption.
NashTech’s wider Leaders Lab research reflects that shift. Around 40% of 52 technology leaders surveyed are still in the AI exploration phase today, yet very few expect to remain there in two years. Instead, leaders expect AI to become embedded in delivery, to scale across their organisations and to generate measurable returns. Security, human accountability and cost are already emerging as prominent concerns.
The clearest theme from the discussion was that AI can move the bottleneck rather than remove it.
Participants described how AI is helping development teams produce more code, but this is creating additional pressure elsewhere in the software development lifecycle. Senior engineers have more code to review, testing teams have more to work through and architecture and governance become increasingly important as the volume of AI-generated code grows.
“As AI capabilities and utilisation have evolved over the last 12 months, the volume of code being generated has increased substantially, which has in turn put considerable pressure on testing functions”.
Peter Truman, Software Development Manager
Grayson Bowler, Head of Architecture, added,
“We’ve sped up that bit in the middle enormously, but it’s the bits either side of it that actually then take all the time.”
For CIOs and CTOs, that changes the question of productivity. Measuring lines of code, development velocity or individual developer output risks missing the bigger picture. If engineering becomes dramatically faster but testing, governance or deployment capacity remains unchanged, organisations have not removed the constraint; they have relocated it.
The priority should therefore be optimising the end-to-end delivery system, not one stage of it.
The working group also highlighted a more strategic workforce question. If AI generates more of the code, what happens to the way engineering capability develops?
Colm Campbell, Head of Engineering, described an emerging issue:
“The review fatigue, the pull request (PR) fatigue from senior developers is an issue... some of the engineers are getting too much code to review.”
That raises questions well beyond productivity. Senior developers traditionally build capability by creating, mentoring and solving difficult engineering problems. If their role increasingly shifts towards reviewing machine-generated output, businesses need to consider engagement and retention, but also how junior developers acquire the experience required to become tomorrow’s senior engineers.
That desire to retain human accountability was shared across the group.
As Paige Hughes, Head of Operations, put it:
“Do we want to hand over that responsibility to AI? Probably not, because we still like to get our hands on it and be that human gate.”
AI adoption therefore cannot be treated simply as a tooling decision. It is becoming an operating model and talent strategy decision.
The discussion also challenged the assumption that greater AI use automatically creates a lower-cost engineering model.
As AI becomes integrated across more of the software development lifecycle, the context required by models increases and token consumption can rise rapidly. Attendees discussed setting usage limits, choosing different models for different tasks and maintaining flexibility between providers.
Colm Campbell summarised the emerging reality particularly well:
“For certain things, it’s still cheaper to do it manually... a senior engineer can get things done more cost effectively than we can with a bot.”
The lesson is not to constrain AI adoption; it is to become more deliberate about it. Technology leaders increasingly need to understand where AI creates genuine economic value, which models are appropriate for which workloads and when human expertise remains the better option.
As experimentation turns into production, governance was another recurring theme.
Participants discussed human checkpoints, technical design authorities, automated testing, model changes and the importance of maintaining accountability for AI-generated outputs. Peter Truman outlined one practical approach to building that confidence – parallel running in production-like environments over several weeks - allowing it to complete the processing in a safe environment while the team monitors its behaviour as expected, before any go-live.
This is an important distinction for technology leaders. Governance should not simply be the gate at the end of an AI project. Done well, it creates the confidence to move faster.
The working group left one message particularly clear: AI-assisted development is no longer simply about giving developers better coding tools.
This has implications for the cost and speed of software delivery, as well as where bottlenecks occur and how teams need to be structured. It also means CIOs and CTOs need to reconsider the skills and responsibilities their teams require, alongside how they manage governance, architecture and intellectual property, and how they retain long-term ownership and control of the technology they build.
NashTech’s latest report describes the same execution challenge: 75% of technology leaders expect AI to have a high impact on software development and 85% expect adoption within two years, while internal expertise, ROI clarity and infrastructure or MLOps maturity remains significant barriers.
The organisations that gain most from AI are unlikely to be those that simply generate the most code. They will be those that redesign the system around it, balancing speed with quality, human expertise with automation and experimentation with the controls required to scale.
A confidential forum for senior technology leaders to share what’s working, challenge assumptions and learn from others navigating similar challenges.
If you’re a CIO, CTO or senior technology leader and would be interested in joining our Leaders Lab, register your interest and join the conversation.