AI-assisted software engineering: Rewriting the build versus buy playbook

When we talk about the rise of AI in the enterprise, the spotlight often falls on customer-facing copilots and conversational agents. But beneath the surface, another transformation is quietly unfolding; one with the potential to reshape how organisations build and deliver software.

Welcome to the age of AI-assisted software engineering. It’s not flashy. It’s not public. But it’s rapidly becoming the most effective and enterprise-ready application of generative AI.

Why start with software development?

Software engineering is an ideal proving ground for generative AI because of four powerful traits:

1. Clear feedback loops – Code either compiles or it doesn’t. Unit tests pass or fail. This binary structure gives AI models rapid, high-quality feedback and enables exponential improvement.
2. Massive training data – Open-source codebases, API docs, and forums like Stack Overflow have created a deep, structured dataset for AI to learn from.
3. Synthetic data generation – AI can “self-learn” by creating new code variations and testing them automatically. Unlike other domains, this self-learning process is not limited by data constraints.
4. Measurable productivity gains – Commits, pull requests, cycle time, and defect rate provide quantifiable evidence of ROI.
   

Put simply: software development is where AI can learn fast, deliver results quickly, and operate within strong guardrails.

It’s already happening

This isn’t theory. Enterprises are already rolling out AI-powered coding tools at scale – and seeing real impact.

• Deloitte has noted that AI-assisted development is reshaping software modernisation economics, lowering the threshold for building bespoke solutions that used to require large teams and long timelines.
• Walmart has reported saving over 4 million developer hours annually through AI-powered development automation. That’s equivalent to 2,000 full-time developers.
• National Australia Bank (NAB) scaled its use of Amazon CodeWhisperer (now Amazon Q for developers) from 20 to over 1,000 engineers in a matter of months, reporting a 40% productivity improvement and 45% boost in code quality. Its New Zealand arm, BNZ, stands to benefit from the same.
  

In these examples, AI isn’t replacing engineers, it’s augmenting them. The best developers are becoming orchestrators: reviewing, refining, and guiding AI-generated code.

“AI-Assisted software engineering has transformed how we do business. Even a year ago we thought that AI would at best be able to build the lego blocks of code but not craft the lego house. Now we are seeing that it can build a lego house from a picture or a brief description. In a year or two from now, the role of the software engineer will be completely transformed again. As software engineers we also need to transform so that we can utilise our craft along with AI to dramatically accelerate our output.”
- Damian Harvey, Deloitte Partner and Software Engineer.

Why CIOs should care now
 

As this capability matures, four strategic implications are coming into focus for technology leaders.

1. Rewriting the build versus buy playbook

For years, the software sourcing strategy was clear: buy standard SaaS and customise sparingly. But AI has shifted the cost-benefit curve. Companies are now using generative AI to reimplement critical systems like CRMs at a fraction of the previous cost. Klarna, for instance, replaced its Salesforce CRM with a GenAI-built internal platform.

What used to take 20 developers to do can now be achieved by five AI-augmented engineers.

For New Zealand, it means Kiwi enterprises could soon start reclaiming their tech stacks from global vendors, building tailored systems that fit local needs and regulatory environments better than any off-the-shelf tool.

2. Scaling context, not headcount

AI changes the scaling model. Instead of “more developers = more output,” the new formula is: “more context per developer = more impact.”

An AI-powered engineer can manage broader domains, absorb legacy systems faster, and focus on complex logic while AI handles boilerplate and migration tasks. The concept of a “10x engineer” is evolving into a “10x AI-augmented team".

Organisations that focus on expanding the reach and decision-making power of each engineer, rather than simply growing headcount, will be the ones who unlock the true scaling potential of AI.

3. Shipping is the new bottleneck

AI accelerates coding, but true speed comes from streamlining the entire path from idea to production. Teams must move beyond siloed roles, adopting value stream-aligned, cross-functional structures such as those described in Team Topologies to continuously refine requirements and respond to change.

Modern delivery demands tight feedback loops, not only in code, but also in product strategy and prioritisation. Value should be measured through outcomes directly tied to customer impact, operational efficiency, and alignment with business goals, not just output volume. While AI can boost delivery metrics, without clear value tracking it risks accelerating the wrong outcomes.

AI has made building faster. The next frontier is making shipping adaptive, meaningful, and predictable.

“True productivity isn’t just about delivering faster; it’s about continuously refining what we build and who we build it for. AI can accelerate output, but real impact comes from aligning teams to value, gathering feedback early, and making sure we’re solving the right problems for the right people.”
- Jane Fitzgerald, Deloitte Partner and Customer Leader.

4. Governing the new code factory

As AI-generated code becomes more common, so do the risks it introduces, ranging from hallucinated logic to subtle bugs and security misconfigurations. This makes human-in-the-loop governance essential, particularly in regulated or high-trust environments.

Organisations must establish clear policies for when to use internal versus public AI models. For example, a bank developing software that interacts with customer data may choose to use a private, fine-tuned model hosted within its own secure cloud environment for generating code that handles transactions or personal information. Public models, while more accessible, may pose risks related to data leakage, regulatory exposure, or lack of control over model behaviour. Clearly defining these boundaries ensures sensitive information never leaves trusted infrastructure.

Additionally, all AI-generated changes should be traceable and auditable, with outputs validated through a combination of automated testing, static analysis tools, and human review. This governance layer is not a blocker to speed, but a safeguard that ensures quality and accountability as AI becomes a core part of the development lifecycle.

“As AI becomes a co-developer, governance will be critical to delivering trustworthy solutions. AI governance is more than just rules; it's about creating a foundation of trust that supports innovation. Imagine it as an essential guide, ensuring AI-generated code acts ethically and reliably while aligning with organisational goals. By keeping humans in the loop, we inject empathy and judgment, crucial for navigating complex ethical landscapes. Governance isn't a barrier; it's a facilitator that empowers us to explore new ideas responsibly, blending human insights with technological advancements to build a future we can all believe in.”
- Aravind Subramanian, Deloitte Partner and AI Advisor

Where to start

For CIOs and engineering leaders wondering how to move forward, here’s a quick playbook:

• Start small and safe: Use AI tools to write unit tests or generate documentation on non-critical systems.
• Invest in fluency: Train your developers in prompt engineering, AI validation, and human-AI collaboration.
• Modernise your DevOps: If AI writes 10x more code, your testing and deployment systems need to handle 10x the volume.
• Update your sourcing lens: Evaluate build-vs-buy decisions with AI augmentation in mind—some things worth buying yesterday may be worth building today.
• Establish governance early: Don’t wait until something breaks. Set clear boundaries and practices for AI-generated code now.

The bottom line

Software development is becoming the first enterprise function truly transformed by AI not because it’s easier, but because it’s the perfect match of structure, feedback, and data.

This is the tip of the spear. Enterprises that embrace AI-assisted engineering now will be faster, more agile, and better prepared for what comes next.

For further information contact our experts, and don’t forget to register for Deloitte’s tech focussed events at Techweek.
 

References

• https://www.deloitte.com/uk/en/Industries/technology/blogs/2024/the-future-of-coding-is-here-how-ai-is-reshaping-software-development.html
• https://www2.deloitte.com/us/en/insights/industry/technology/how-can-organizations-develop-quality-software-in-age-of-gen-ai.html
• https://www.youtube.com/watch?v=mLV82tpc6jQ