While some engineering teams are implementing systematic AI agent frameworks, other teams are accumulating AI tools. The difference? The teams deploying strategic frameworks see 25% improvements in velocity and quality of code improvements, in addition to better engineer adoption.
The cost of unstructured adoption
We are now at a crossroads where developers are using AI tools more than ever. Yet this “code explosion” can drive increased maintenance, increased refactoring, and debugging more difficult if done haphazardly. At Nearform we’re experimenting using agents in new and mature software projects. As we have learnt along the way we can now pull back the curtain on lessons learnt to date.
Most teams are still at the early stages of assessing tooling and experimenting with a wide variety of applications without strategic integration. This works for evaluation but as we mature, if this approach is left unchecked, it can create a chaotic ecosystem where different developers use different AI systems for similar tasks. The result? Automated inefficiency with inconsistent outputs, conflicting code patterns, and increasing technical debt – the very opposite of what was promised.
Frameworks that scale with engineering best practice
Nearform's 14+ years of Software Development Life Cycle (SDLC) expertise indicates that the future isn't about having more AI tools — it's about implementing scalable frameworks and ways of working that help AI agents work together systematically alongside humans. This prevents quality degradation that comes from inconsistent AI system usage. It also allows for an iterative slow release and test deployment.
AI systems may be new, but fundamental engineering practices are tried and tested. Experience in building robust systems, maintaining code quality, and implementing systematic processes becomes even more valuable when managing AI agents that can amplify both good and bad practices at unprecedented speed. In short, there is a cascading and amplifying effect once things go wrong with an LLM.
Is ‘role-based engineering’ a possible solution?
Role-based engineering isn't about giving AI tools better prompts. It's about creating AI agents that function as specialised team members with distinct responsibilities, expertise areas, and collaborative protocols.
Consider some of the members of a typical development team structure:
- Human senior developer: Focuses on complex business logic, architectural decisions, and mentoring
- AI testing agent: Automatically generates comprehensive test suites, maintains test coverage, and identifies edge cases
- AI documentation agent: Maintains up-to-date system documentation, generates API specs, and creates onboarding materials
- Human product manager: Defines requirements, manages stakeholder communication, and makes strategic decisions
- AI UX pattern agent: Ensures design system consistency, validates interaction patterns, and maintains component libraries
Each AI agent in the example above has its own domain expertise, operating parameters, and quality standards. They don't just respond to requests — they proactively contribute to the development process within their areas of responsibility. More importantly they have defined goals, permissions and responsibilities that can be checked and validated.
Hybrid teams, not human replacement
Nearform’s approach to AI agent integration emerged from a simple recognition: successful AI implementation is a collaborative process. There’s considerable media hype about AI replacing human expertise. Yet instead, Nearform clients see success in creating hybrid teams, where AI agents and seasoned developers collaborate as colleagues, each contributing unique strengths removing some tedious or repetitive tasks.
During an “Ignite”, our project discovery process, Nearform's proprietary “market intelligence agent” conducts market research, competitive analysis, and product requirement research, while senior designers and technical directors focus on business alignment and stakeholder management. This shortened a process from days to just hours, allowing greater human engagement while optimising the more mundane research tasks.
After discovery, “AI development agents” handle scaffolding, boilerplate generation, and pattern implementation while human developers focus on complex business logic and system integration. “AI testing agents” maintain quality assurance and regression testing while human QA engineers design test strategies and handle edge cases. This is governed by an overarching framework that supports a sequential process - always involving a human handoff.
For older legacy systems, AI agents also handle documentation updates, minor security patches, and performance monitoring while human engineers focus on system evolution and strategic improvements. This drastically improves onboarding and knowledge transfer to new engineers.
Nearform’s battle-tested approach is based on a core understanding: AI agents excel when given clearly defined objectives, responsibilities and permissions. Playing to their strengths of thoroughness and pattern recognition is key, while humans can excel at creativity, strategic thinking, and complex problem-solving. Role-based engineering harnesses both, allowing the best people to do more valuable work.
Stay tuned for the rest of Nearform’s series on AI in the software development lifecycle.
But wait - there's more.
Nearform publishes real-world learnings on data & AI, engineering, and digital strategy - with more merged in weekly.
Insights
Perspectives on AI in engineering, product development, and strategy, for enterprise executives.
Community
Deep dives and tutorials by engineers, for engineers.
You may also like
OpenAI goes open source: Nearform experts share how enterprises can benefit - Part 1
