AI-assisted engineering is only one part of the story. Strong teams are also investing in verification, platform foundations, modernisation and secure delivery.
1. AI becomes part of the delivery system
AI coding tools have moved from experiment to everyday workflow. Google’s 2025 DORA research surveyed nearly 5,000 technology professionals: 90% reported using AI at work and more than 80% believed it increased productivity. But 30% reported little or no trust in AI-generated code. The trend is therefore not simply 'more generated code'; it is the redesign of review, testing and documentation around AI-assisted work.
DORA describes AI as an amplifier. It helps strong systems move faster, but can also magnify weak feedback loops and increase delivery instability. Teams that benefit most pair AI assistance with small changes, automated tests, clear architecture and fast human review.
2–4. Agents, retrieval and modern platforms
AI agents are moving beyond chat to execute multi-step tasks across software and business workflows. McKinsey’s 2025 research found 23% of surveyed organisations were scaling an agentic AI system somewhere in the enterprise, while another 39% were experimenting. Useful implementations keep permissions narrow, log actions and require human approval for consequential decisions.
Retrieval-augmented generation (RAG) remains important because businesses need answers grounded in their own controlled information. At the same time, API-first modernisation and internal platform engineering are making legacy data safely available to new applications and AI services. The competitive advantage often comes from connecting systems and improving workflows—not from the model alone.
5–6. Verification and security move to the centre
As producing code becomes faster, review and verification become the constraint. Teams are investing in automated acceptance tests, dependency controls, observability and evaluation of AI outputs. The question changes from 'How quickly can code be written?' to 'How confidently can a useful change reach production?'
Security must also account for non-human actors. Agents need identities, least-privilege access, audit trails and limits on the tools and data they can reach. This makes secure architecture, data governance and operational monitoring product requirements rather than late-stage compliance tasks.
Key takeaways
- AI-assisted development is mainstream, but trust and stability remain open challenges.
- Agentic systems need narrow permissions, observability and human escalation paths.
- RAG and APIs connect AI to useful, governed business context.
- Testing, review and security become more valuable as code generation accelerates.