I can see a paradox in the software development industry. A rigorous 2025 study found that when developers use AI-tools, they take 19% longer to complete tasks – contradicting developers’ belief and expectations that AI tools would speed them up by 24%. Yet, 90% of software development professionals now use AI tools, with a significant majority reporting heavy reliance. Over 80% indicate that AI has enhanced their productivity, and 59% report a positive influence on code quality (2025 DORA Report).
This paradox reveals a deeper truth: AI-assisted development isn’t simply about making existing developers faster – it’s about transforming the roles that software development participants play. The most critical factor in AI tool speed and productivity is the quality of functional requirements and feature specifications provided to the AI, which elevates the importance of Product Managers in making AI-driven development successful.
In traditional software development, a Product Manager typically writes a brief feature description, and let developers to figure out implementation details. With AI-assisted development, this dynamic has fundamentally shifted. Product Managers can now collaborate with AI to create robust requirements and then let Claude Code or Cursor to generate the code off these specifications and test it, Then submit implemented feature for review by senior developers or architects without any intermediate step required.
The message is clear: those who write better requirements, design better systems, and validate more effectively will thrive. Those who simply write code from specifications face an uncertain future.
Understanding the Productivity Paradox
When AI coding assistants first emerged, the promise seemed straightforward: developers would code faster, ship more features, and have time for higher-level work. Reality proved more nuanced.
How do we reconcile the contradictory findings described earlier? I think that the answer lies in understanding that AI’s impact varies dramatically based on:
- Developer experience level: Senior developers have experience in explaining others, including AI assistant how to implement solution
- Task complexity: Simple boilerplate coding typically does not need as much explanation as that required for complex architectural decisions
- Most critically: The quality of product requirements and feature specifications provided to the AI assistant.
This last point is where the real transformation is occurring.
The Rise of Requirements-Driven Development
Product Managers: The New Critical Role
In traditional software development, a Product Manager often asked by engineering to write only a brief feature description, and let them to figure out the implementation details. With AI-assisted development, PM can write detailed functional requirements, ask AI to propose implementation plan based on these requirements, review the plan, make suggestions, and then, after couple iterations, ask AI assistant to implement the code, test suite, and run the tests.
Dynamic has fundamentally shifted – after requirements is written, it takes a few hours to get new feature implemented and ready for review by a senior developer.
Here is an example from Google’s senior director of product management for developer tools describes using AI to create specification documents: “I will use Gemini CLI in order to create a more robust requirement doc in Markdown. That will usually create probably about 100 lines of fairly technical, but also outcome-driven specification. Then I will use Gemini CLI to write the code based on that specification”. I have similar experience with Claude Code.
The pattern is clear: better specifications yield exponentially better AI-generated code. Product Managers must now develop proficiency in low-code platforms:
- Requirements Documentation: Creating detailed, technically precise specifications that serve as effective AI prompts
- Data Literacy: Understanding how AI processes information and what context it needs
- AI Tool Selection: Evaluating which AI coding assistants best fit project requirements
- System Design Thinking: Translating business needs into architectural concepts AI can implement
Research analyzing 500,000 coding interactions found that 79% of Claude Code conversations were classified as automation, where AI directly performs tasks, versus augmentation where AI collaborates with humans. This automation-heavy usage underscores the critical importance of well-specified requirements—when the AI is doing most of the work, the quality of instructions becomes paramount.
The shift toward requirements-driven development creates a new documentation paradigm. PMs must now maintain:
- Comprehensive feature specifications: Not just “what” is required, but detailed “how” at a technical level
- Context documents: Business logic, user workflows, edge cases, and constraints
- Integration requirements: How features connect with existing systems
- Success criteria: Measurable outcomes and validation approaches
This isn’t traditional PM documentation—it’s technical specification writing that directly feeds AI coding tools.
The Evolving Role of Software Architects
As a Product Manager myself, I was focusing more on the impact that advent of AI-assisted software development has on the role that Product Managers should play in this process. I think though that AI uptake brings an eminent changes into the role that Software Architect, DevOps and QA will play in implementation and deliver of reliable, stable, secure and performant software systems. Let’s start with Software Architect role.
While Product Managers define what to build, senior developers / software architects determine how to build it. In the AI era, this distinction becomes even more important.
From Code Design to AI Orchestration
I think that software architects will not be replaced by AI any time soon, but they will be replaced by software architects who know how to leverage generative AI, LLMs and agentic workflows in their applications. And just as importantly, know how and when NOT to use generative AI.
There is a paradox. I think that AI cannot make architects obsolete because LLMs are pre-trained on existing datasets – they excel at producing architectures they’ve seen before and generating solutions that are already built. But AI would struggle with unique, context-specific architectural challenges. Therefore, core responsibilities of architects will remain mostly the same in the AI era:
- Assessing Technologies: Selecting appropriate software technologies, AI frameworks, and tools for projects, then overseeing integration of AI systems with existing infrastructure and applications
- Ensuring Non-Functional Requirements: Guaranteeing scalability, performance, security, and maintainability of AI-enhanced solutions
- Context Translation: Understanding business context and organizational factors that aren’t encoded in code i.e., doing work that AI cannot replicate
- Human Element Navigation: Managing situations where key solution aspects like acceptable quality, latency, and cost often lack clarity during scoping and design
The Human Element in Architecture
Software architecture requires uniquely human skills: effective communication and understanding how different stakeholders think. AI cannot replicate these abilities.
AI can assist with five of six core architectural activities—clarifying requirements, designing structures, documentation, evaluation, and implementation support. However, stakeholder communication and understanding organizational context remain distinctly human responsibilities. Hence, architects must be able now:
- Generate proof-of-concepts using AI tools rapidly
- Validate AI-suggested architectures against real-world constraints
- Translate between business stakeholder language and technical AI specifications
- Identify where AI-generated solutions can fail and custom design is required
- Create architectural guardrails for AI-generated code
Summing up, architect role becomes more strategic, more communication-intensive, and more focused on judgment rather than detailed architecture design
DevOps and QA: The Quality Gatekeepers
As AI generates more code and much quicker, the importance of code validation, testing, and deployment governance increases exponentially.
DevOps in the AI Era
AI coding tools enable automated test generation whereby LLM algorithms analyze code changes and automatically create test cases. This drastically reduces the need for manual test creation and as a result, helps with speeding up CI/CD pipelines. Teams using AI coding assistants complete feature implementations 41% faster according to GitHub’s 2025 State of the Octoverse report.
But speed without quality creates problems. DevOps and QA professionals now face expanded responsibilities:
- Testing Strategy and Automation:
- AI automatically generates and runs test cases based on code changes, improving test coverage while reducing manual effort
- Self-healing systems use AI to detect anomalies and automatically resolve issues without human intervention, minimizing downtime
- AI-powered anomaly detection in CI/CD workflows catches problems before deployment, with real-world projects showing 96% accuracy rates, 87.5% precision, and 100% recall scores
- Security Integration (DevSecOps):
- Adopting shift-left testing – AI can run vulnerability detection earlier and catch security issues while code is being written rather than waiting until predeployment reviews by security teams. With AI tools generating code rapidly, catching security issues immediately becomes critical to prevent massive backlogs of security reviews
- Using AI for threat detection – organizations doing that can identify and contain breaches on average, 108 days faster according to IBM’s 2025 Cost of a Data Breach report
- Continuous security scanning of AI-generated code for vulnerabilities and compliance issues
- Predictive Operations:
- AI can forecast potential system outages or performance bottlenecks by analyzing historical trends, telemetry data, and contextual signals
- Automated rollbacks when anomalous post-deployment behavior is detected, protecting end users from experiencing issues
- Optimization of deployment timing based on system load and business impact
The Test Architect Role
A new role – Test Architect – has emerged as a leader responsible for designing, implementing, and overseeing testing frameworks. Test Architects are taking a holistic approach to ensure testing strategy aligns with project goals, architecture, and delivery timelines.
Unlike traditional QA engineers who execute test cases, Test Architects:
- Design comprehensive testing frameworks that handle AI-generated code volume
- Ensure testing covers both functional and non-functional requirements
- Advocate for shift-left testing approaches
- Integrate AI-driven testing tools while maintaining human oversight
Role Evolution: From Execution to Strategy
The DevOps and QA role transformation mirrors the broader shift: from hands-on execution to strategic oversight. DevOps and QA professionals now:
- Define testing strategies rather than write individual tests
- Configure AI testing tools rather than manually test features
- Monitor system health through AI analytics rather than reactive troubleshooting
- Design deployment governance rather than manually deploy releases
- Validate AI decisions rather than make every operational decision
This elevation of the roles requires new skills: understanding AI capabilities and limitations, designing effective automation, and knowing when human judgment remains essential.
Conclusion: Evolution, Not Extinction
The software development profession isn’t ending—it’s transforming, just as it has with every major technology shift. With advent of AI-assisted software development I can see the following four roles becoming more critical:
- Product Managers
- Create detailed technical specifications that guide AI code generation
- Bridge business requirements and technical implementation
- Select and configure AI development tools
- Define success criteria and validation approaches
- Manage the expanded “software creation ecosystem” where AI assists multiple stakeholders
- Software Architects
- Design systems that AI cannot create from “cookie-cutter” templates
- Provide organizational and business context AI lacks
- Select appropriate technologies and integration approaches
- Ensure scalability, security, and maintainability
- Validate AI-suggested architectures against real-world constraints
- DevOps and QA Engineers
- Define testing strategies for high-volume AI-generated code
- Implement automated validation pipelines
- Ensure security through shift-left practices
- Configure and monitor AI-driven deployment systems
- Validate AI decisions in production environments
- Senior Engineer / Dev Managers
- Review and validate AI-generated code for quality and correctness
- Provide domain expertise and contextual knowledge
- Make architectural decisions requiring human judgment
- Mentor teams in effective AI tool usage
- Maintain code standards and technical culture
At the same time, the shift toward AI-assisted coding isn’t about replacing developers; they must evolve into overseers of quality, security, and maintainability – becoming architects, reviewers, and orchestrators of AI-driven development. Those who write better requirements, design better systems, and validate more effectively will thrive.
Igor's Musings About Technology 