AI-Powered Coding: Unlocking Billions in Savings for Public Sector Tech

The AI revolution in software development is not about generating chatbots or fancy Japanese anime images; it's about fundamentally changing how code and software is created and maintained.

AI-powered tools are accelerating development times and boosting developer productivity like never before. It has even created a phenomenon called vibe coding to reflect a change in some programmers' role from manual coding to guiding, testing, and refining the AI-generated source code.

However, the potential of these advancements to significantly reduce technology spending within the public sector remains largely untapped. This is critical because in Government, we all need to think about more efficient use of funding and more effective delivery.

This article outlines the massive savings potential and provides actionable steps for government leaders, procurement specialists, and ICT professionals to leverage AI-driven efficiencies and ensure taxpayer dollars are used more effectively.

The AI-Powered Development Revolution:

We've moved far beyond simple code completion. Today, developers are leveraging sophisticated AI tools and frameworks (often in combination) like GitHub Copilot, Cline, Cursor AI, Windsurf, Bolt.new and Bolt.diy to:

• Automate Code Completion and Generation: Significantly reduce manual coding effort with intelligent suggestions and generation.

• Enhance Code Understanding: Chat with your codebase to quickly understand complex logic, identify potential issues, and generate documentation.

• Streamline Code Modification: Update multiple files simultaneously, apply consistent changes across projects, and refactor code with AI assistance.

• Improve Code Quality: Use AI to identify potential bugs, security vulnerabilities, and performance bottlenecks early in the development process.

• Optimise Navigation: Quickly find relevant code sections, understand dependencies, and navigate complex projects with ease.

  
  

Seeing is believing: Here is a short video of me using Cline, the leading AI coding assistant, to do some coding in VS Studio Code a popular IDE (73.6% of coders use Visual Studio Code - 2024 Stack Overflow Developer Survey) – sped up for non technical users so I could embed it as a video.

What that means is I can use natural language to code, and the AI assistant powers through creating the code /scripts saving me time (although I still need to review ti).

Note: The underlying LLMs / multi-modal foundation models (MFMs) are continually improving on coding benchmarks/tasks such as the Aider’s polyglot benchmark and the SWE-bench Verified for broader tasks. The past two years of progress in coding benchmarks is a strong indicator that AI will play an increasingly capable (and collaborative) role in software development going forward.

Add to this, behind the scenes, lesser-known breakthroughs in infrastructure and protocol design are also quietly driving this shift. Beyond AI-powered IDEs, breakthroughs in model content protocols and runtime performance libraries are dramatically boosting development velocity. Libraries like UV, the ultra-fast Python runtime built in Rust, are shaving seconds off every iteration. Meanwhile, emerging standards like Model Content Protocols help streamline giving agents' tools.

---

The Evidence

Technical Analysis of AI-Driven Productivity Gains

Recent studies have quantified the productivity benefits of AI-powered coding tools across different phases of the software development lifecycle:

• Code Documentation: 45-50% faster completion rates (McKinsey study)

• Code Generation: 35-45% productivity improvement

• Code Refactoring: 20-30% time savings

• Overall Code Quality: 20-40% improvement in Greenfield projects

In controlled experiments, developers using AI assistance have demonstrated remarkable efficiency gains:

• A Cornell University study involving 95 programmers tasked with implementing an HTTP server in JavaScript found that participants with GitHub Copilot access completed the task 55.8% faster than the control group.

  
  

These productivity gains are most pronounced when multiple complementary AI tools are used together, each focusing on different aspects of the development workflow.

The Financial Implications for the Public Sector:

Governments worldwide are investing heavily in technology. Consider these figures:

• Australia: Projected to spend over AUD 19 billion on IT in 2024 (Source: ITNews).

• English public sector: Public sector spent approximately £17.3 billion on IT in 2022 (Source: Tussell).

• United States: For the 2024 fiscal year, the U.S. federal government allocated around 30.7 billion U.S. dollars for major federal IT investments. The total amount of spending on IT is expected to amount to over 102 billion U.S. dollars in 2024, significantly higher than the federal government's IT budget in the previous year (Source: Statista).

Even with conservative estimates of productivity gains from AI-powered coding tools, the savings potential is substantial:

Country	Annual IT Spending	10% Savings	20% Savings
Australia	AUD 19 billion	AUD 1.9 billion	AUD 3.8 billion
England	£17.3 billion	£1.73 billion	£3.46 billion
USA	USD 30.7 billion	USD 3.7 billion	USD 7.2 billion

Based on McKinsey's analysis, AI could impact current spending on software engineering functions by 20-45%, suggesting that our table may actually underestimate the potential savings.

Specific AI Applications Across the Software Development Lifecycle

Inception and Planning

• Natural language processing of requirements documents

• Automated user story generation

• Stakeholder communication enhancement

• Rapid prototype generation

Design Phase

• Generation of multiple architecture alternatives

• Early research analysis

• Virtual design simulations

• Design review regulation compliance

• API design optimisation

Development Phase

• Automated code generation

• Code translation and migration

• Intelligent code completion

• Cross-language support

• Documentation generation

• Bug prediction and prevention

Testing and Quality Assurance

• Automated test case generation

• Test data set creation

• Unit test writing

• Regression testing optimisation

• Security vulnerability detection

• Performance bottleneck identification

Maintenance and Support

• Legacy code understanding

• Modernization assistance

• Bug triage and prioritisation

• Knowledge base maintenance

• Support ticket analysis

How to Capture These Savings: Actionable Strategies for Government:

To unlock these savings and ensure they benefit the public, governments must take proactive steps:

• Educate Leadership: Educate leaders and procurement personnel about the untapped potential and practical applications of AI-powered coding tools. Many still view AI conceptually without understanding its concrete applications.

• Embrace Outcome-Based Contracting: Shift from paying for hours worked to paying for clearly defined results. This incentivises vendors to maximise efficiency (Source: OECD).

• Implement Productivity-Indexed Pricing: Adjust payments to vendors based on demonstrated productivity gains achieved through the adoption of AI tools. Establish clear metrics and verification processes.

• Mandate Technology Transfer: Require vendors to use AI-powered coding practices and to train government staff on these tools. This builds internal capacity and reduces long-term reliance on external contractors.

• Utilise Shared Savings Contracts: Structure contracts that incentivize vendors to share a portion of the cost reductions achieved through increased productivity.

• Demand Transparency in Procurement: Explicitly require vendors to demonstrate how AI will deliver specific, measurable cost savings in software development. Scrutinize proposals to ensure these savings are genuine and not offset elsewhere.

• Address Implementation Challenges: Implement clear governance frameworks addressing intellectual property concerns, privacy protection, security requirements, and code quality standards when adopting AI coding tools.

Market Growth and Future Trajectory

The market for AI-powered development tools is projected to grow at a compound annual growth rate (CAGR) of 21.4% between 2022 and 2032. This growth trajectory indicates both the technology's increasing capability and its broader acceptance within enterprise and government sectors.

Current market analysis predicts that by 2032, the global AI in software development market will reach approximately $169.2 billion USD, up from just $25.4 billion in 2022. This expansion is primarily driven by investments in:

• Code generation capabilities

• Code optimisation tools

• Bug detection systems

• Testing and quality assurance automation

Crucial Considerations and Caveats:

·         Holistic Approach: Coding productivity is only one piece of the puzzle. Consider the entire project lifecycle, including architecture, integration, testing, security, and user experience.

• Realistic Expectations: Be realistic about the potential impact of AI and focus on measurable results. Many software engineering activities will still require significant human expertise and context.

• Transparency is Key: Openly assess and share the savings from AI-driven productivity gains with stakeholders and the public.

• Skill Development: Invest in training for prompt engineering and AI tool utilisation to maximise productivity gains.

• Collaboration Enhancement: Recognise that AI tools can transform team dynamics by facilitating knowledge sharing, enabling natural language interfaces to code, and improving code review processes.

• Code Quality and Security Concerns: AI-generated code may still contain bugs or security vulnerabilities. Human oversight and rigorous testing remain essential.

• Intellectual Property Risks: Many commercial GenAI tools have been trained on open-source (and potentially proprietary) codebases, raising legal questions about IP rights and licensing compliance.

• Uneven Benefits Across Projects: The productivity gains will not be uniform. Complex, mission-critical systems may see less benefit than more straightforward applications.

• Initial Investment Requirements: Adopting AI tools requires upfront expenditure on licenses, training, and workflow adjustments before savings are realised.

• Dependency Concerns: Over-reliance on AI tools may reduce developers' fundamental skills over time or create dependency on specific vendors.

A Balanced Approach

• While AI-powered code editors/tools promise substantial efficiencies, governments must remain realistic. Coding productivity and accelerated software development improvements represent just part of broader project delivery processes, which also include architecture, integration, testing, security, and user experience considerations.

• Some projects may see dramatically higher productivity gains than others. System complexity, security requirements, team experience with AI tools, and integration demands all affect the real-world benefits achievable. Moreover, public sector projects often have unique constraints—including legacy system integration, accessibility requirements, and enhanced security needs—that may limit AI tools' effectiveness without careful adaptation.

• Public sector organisations should start with pilot projects, measure outcomes rigorously, and scale successful approaches methodically. This measured approach helps manage expectations while building institutional knowledge about how to deploy these tools effectively.

However, let me be clear: the mega-trend we are seeing is that the use of AI in the ideation, build and deployment of these technological solutions is accelerating...it isn't slowing down.

Call to Action:

• The AI revolution in software development presents a unique opportunity for governments to significantly reduce technology spending and reinvest those savings into critical public services. By embracing AI-powered coding tools, reforming procurement processes, and demanding transparency from vendors, public sector organisations can unlock billions in savings and deliver better outcomes for citizens.

• We must approach this opportunity with both enthusiasm and caution—implementing AI tools strategically, measuring outcomes honestly, and adjusting our approach based on evidence rather than hype. With careful planning and appropriate governance, these technologies can help us deliver more value to citizens while reducing costs.

• Let me know your thoughts. And the steps you will take to champion the adoption of AI-powered coding in your organisation? If you are interested in this blog, reach out to me.

As someone who builds AI solutions and works in the public sector, I am passionate about passing on these savings to help fund other crucial areas.

---

Defining the Public Sector: The public sector, defined as including any governmental organisation at federal, state, county, or local levels, intergovernmental agencies (e.g., NATO, the EU, OECD, UN), and entities such as policing, utilities, public transport, infrastructure, and certain universities. Thus, public sector technology investments profoundly affect a broad range of vital societal services.

Technical note: All AI-powered tools focused on accelerating software development, though they approach this goal in slightly different ways, for example:

• Cline: An AI-powered command-line interface that lets developers interact with AI coding assistants directly from their terminal.

• Cursor AI: Specifically, an AI-enhanced code editor that integrates AI capabilities directly into the development environment.

• Bolt: While primarily an AI acceleration framework, it's used in development workflows to make AI-powered coding tools respond more quickly.

• Windsurf: Though focused on model training optimisation, its primary application is in building more effective code generation and assistance tools

Vibe coding: Vibe coding (also vibecoding) is an AI-dependent programming technique where a person describes a problem in a few sentences as a prompt to a large language model (LLM) tuned for coding. The LLM generates software, shifting the programmer's role from manual coding to guiding, testing, and refining the AI-generated source code.  Vibe coding is claimed by its advocates to allow even amateur programmers to produce software without the extensive training and skills required for software engineering. The term was introduced by Andrej Karpathy in February 2025 and listed in the Merriam-Webster Dictionary the following month as a "slang & trending" noun. (Source : Wikipedia)

Author: Christopher Foster-McBride, CEO of AI start-ups and Public Sector Digital Lead