Development

min read

The net value of AI in software delivery

Rich Atkinson

June 2, 2026

Overhead view of a software engineer working at a laptop, keyboard and mouse on a wooden desk.

AI accelerates some software delivery work. But does better productivity mean better software?

For custom software delivery teams like us, it’s less about whether we can change code quickly and more about whether we trust a change before it reaches users.

Imagine a software engineer is working on a new release for a client’s food-ordering app. Before the release goes live, an automated test in the order flow fails. The software delivery team checks if the app still enables users to select items from the menu, complete payment, and successfully place an order. If the team discovers users can’t make payment or place an order, it needs to work out what needs to be fixed first.

An AI coding assistant like GitHub Copilot may be able to suggest a fix within minutes. But the engineer still needs to check how the suggested fix changes the codebase, and whether it affects product behaviour.

Experienced engineering judgement is needed to decide if the suggested fix solves the actual order-flow issue, without creating more issues.

‍

How AI is changing expectations for software delivery?

As our whitepaper, The Business of AI, highlights, AI is having a moment like the Internet did in the 1990s. Its mainstreaming is evident in the vast number of work tools and platforms it now lives in. In our software delivery workflows, we’re discovering an increasing number of use cases.

Stack Overflow’s 2025 Developer Survey – which is part of a growing body of data illustrating the widespread adoption of AI tools in software delivery and development – found 84% of its surveyed developers are using or planning to use AI tools in their development process, and 51% are using them daily.

The organisations we speak to are becoming more curious about what AI can do for software delivery performance, timelines, and cost.

It’s clear that AI has the potential to help software teams reclaim time and compound productivity, but its net value is commensurate with the engineering capabilities of the teams who use it.

‍

Where AI is proving useful in software delivery

Our AI whitepaper sets out a wide variety of use cases for AI in software delivery today. Software development and design teams harness AI for exploring complex problems, receiving code assistance, generating throwaway code for prototyping and exploration, and generating technical documentation.

Sonar’s State of Code Developer Survey, a survey of more than 1100 developers, found its respondents are using AI for prototypes and proofs of concept (88%), internal non-critical production software (83%), customer-facing applications (73%), and business-critical or mission-critical services (58%).

In our conversations with our developers, we are talking about how they are using AI tools for software delivery activities including code generation, rapid prototyping, code review support, bug investigation, documentation and exploring implementation options.

Our engineers are developing their work processes with tools such as Claude, GitHub Copilot, OpenAI tools including Codex, Gemini, and NotebookLM.

Some of the common observations our team are making are that AI can improve how they ideate. In finishing a prototype in a fraction of the time, engineers can test whether an approach works before they commit more delivery time to it. An AI-assisted code review can flag issues the reviewer may want to investigate further. An AI-assisted bug investigation can help identify where a problem is likely to be coming from. Engineers can spend less time settling on the starting point, and more time deciding the right fix.

A simple calculus for AI-assisted delivery

AI can make a software task faster, but that does not automatically make the whole project faster or cheaper. The real calculation is not whether AI produced a code draft, prototype, or suggested fix faster than a human could. The calculation is what happened after that.

If the AI-assisted work is accurate, well-structured and easy for engineers to check, it may reduce total effort. If it creates unclear code, misleading fixes, extra review work or technical debt, the time saved upfront might be lost later.

Organisations assessing AI’s usefulness in software delivery need to measure more than how quickly it produces the first version of the work. They need to know whether the software team can turn that first version into production-ready software without creating extra work elsewhere in the delivery process.

Managing the risks

Quickly generated AI work isn’t inherently “slop”. Nor is it a panacea. The more skilled the engineering team, however, the better they can see it for what it is — the good and the bad.

One of the risks software delivery teams need to manage is the uncanny valley of AI code.

In just a glance, things might look pretty right — but something’s not quite right. The bloated code, hallucinations, and logic errors might be quite insidious.

As developer surveys show, that means more (significantly more, sometimes) review, testing and correction work for the engineers responsible for checking it.

AI can miss context during refactoring and test generation because it lacks full architecture awareness and team coding conventions.

We also know that AI can add unnecessary complexity in longer-term application work. When it solves one local problem, it might create complications for other parts of the system.

How Airteam manages responsible AI use

Our AI policy includes clear rules that define where and when AI tools can be used
AI-assisted work is reviewed before it becomes part of the product
We protect sensitive client data by ensuring it’s excluded from AI tools, with mock data used where needed
We use zero-logging tools where appropriate, so code is not stored by the provider
Our engineers continually assess tools, as models, interfaces and platforms change.

Keep in mind, the same AI platform can produce different results for different software teams, depending on how clearly the engineers frame the task, share context, and review the results, so risk management is also about developing AI skills to get the best out of tools that are always evolving.

We’re invested in discovering where the tools help and where they hinder (more on that later). We also manage AI-assisted engineering work through the quality and security disciplines that guide our broader software delivery work. We focus on senior engineering review, secure development practices and ISO-certified management systems.

‍

AI’s potential is better software delivery, not just producing code

Ultimately, we’re more interested in continuing our discovery of how AI helps us deliver better-quality software overall.

For organisations investing in digital products that can win or lose favour with users, it is important to look for a software delivery partner that can offer more than speed.

Did you know?

Airteam has been building and using AI tools internally for some time. It has also been steadily upskilling its team in how to use the technology responsibly and effectively.

That work includes client-facing AI projects across different use cases, including a sales training tool, an AI-supported healthcare service tool, and early exploration of product features where AI helps users customise content while traditional software controls handle the parts that need to be structured, accurate and verifiable.

Airteam’s AI capability is also being developed at a leadership level. Our Executive Director Patrick Goffin has completed the AI Fluency for Directors program through the Australian Institute of Company Directors and the University of Sydney.

Airteam’s white paper, The Business of AI, also reflects the team’s early thinking on where AI can create practical business value without being added for its own sake.

The larger opportunity is not just faster code, but better delivery decisions across the whole project.

We currently have semi-regular catch-ups to discuss the tools we’re using and the discoveries we’re making, and we encourage our team to try different tools – and in different contexts.

As AI-assisted delivery matures, we’re looking forward to seeing how we will invest our time and expertise into decisions that continue to enrich software quality for the organisations we partner with.

Two software developers working on laptops during a collaborative coding session.

Reach out to Airteam

If you’re exploring how AI should shape your next software project, speak with Airteam about custom software development that uses AI pragmatically, with appropriate oversight, security considerations and delivery accountability. Reach out via our contact form or email us directly at hello@airteam.com.au

‍

Questions to ask your software delivery partner about AI

How is AI being used in the development process?
A software delivery partner should be able to explain where AI is used, such as prototyping, code drafting, review support, bug investigation and failed test analysis.

Which AI tools are being used?
A software delivery partner should be able to name the tools being used and explain why different tools are suitable for different tasks. Popular tools engineers use include Claude, GitHub Copilot, OpenAI/Codex, Gemini and NotebookLM.

Does AI make software development cheaper?
AI can reduce effort in some tasks, but that does not automatically reduce the overall cost of delivery. The work still needs to be reviewed, tested, integrated and maintained by experienced engineers. AI should be used to add value to software delivery.

Does AI reduce the number of developers needed on a project?
AI can reduce effort in some tasks, but complex software still needs experienced engineers for architecture, security, testing, maintainability, and delivery accountability.

Can AI write most of the code now?
AI can generate code quickly, but that code still needs to be understood, tested, refactored and fitted into the wider product.

How is AI-generated code reviewed or validated?
Engineers review, test and rework AI-assisted code before it becomes part of the product.

How are security, confidentiality and IP protected?
Sensitive client data should not be entered into AI tools, and appropriate privacy settings should be used where code is involved. Airteam also holds ISO 27001 and ISO 9001 certifications, supporting its information security and quality management practices.

Where can AI create risk or rework?
AI can create technical debt, verbose code, misleading fixes or extra review work if engineers do not check it properly.

What work still requires experienced human judgement?
Architecture, security, quality assurance, maintainability, risk management and delivery accountability still need experienced engineers.

Who remains accountable for the final software delivered?
The software delivery partner remains accountable for the product, even when AI assists with parts of the work.

Share this post