As AI agents become increasingly capable of writing software, developers will still have a critical role to play in ensuring that the resulting code is reliable, resilient and fit for production, according to Darko Mesaroš, Distinguished Developer Advocate at Amazon Web Services (AWS).
Mesaroš said developers should continue to strengthen their technical skills rather than relying entirely on autonomous AI tools.
“I always encourage fellow developers to still keep honing their craft as they really know what they are doing and an agent is only going to make them more productive,” he said while speaking to the media.
His comments come as AI-assisted and agentic software development rapidly becomes part of mainstream development workflows, while concerns over low-quality AI-generated output, often described as “AI slop,” continue to grow.
See more of our coverage in your search results.
Add INDYASTORY on GoogleWhy Developers Still Matter in the Age of AI Agents
AI agents can generate code, modify existing software and carry out increasingly complex development tasks. But Mesaroš pointed out that software development does not end when code has been generated.
More Read
Applications can encounter unexpected problems once they are deployed. Operational metrics can change, requirements can evolve and systems can behave differently from what developers initially anticipated.
These situations require people who understand the underlying technology and can determine what needs to change.
For Mesaroš, that means the developer’s role is evolving rather than disappearing.
More Read
Instead of focusing exclusively on writing every line of code manually, developers increasingly need to direct, review and orchestrate AI agents while maintaining responsibility for the final software.
See more of our coverage in your search results.
Add INDYASTORY on GoogleWhat Is AI Slop?
“AI slop” generally refers to low-quality, generic or poorly reviewed content generated with AI.
In software development, the concern is that developers could use AI tools to generate large amounts of code without adequately checking whether that code is secure, maintainable, accurate or appropriate for the specific application.
Mesaroš argued that developers need to ensure their projects are resilient to this type of output.
See more of our coverage in your search results.
Add INDYASTORY on Google“The responsibility of a developer is to make sure that the project they’re building is resilient to AI slop,” he said.
That places greater importance on software architecture, testing, code review, security and understanding the requirements behind a product.
Developers Are Becoming AI Orchestrators
The rise of AI coding agents is changing the nature of software development.
Developers can increasingly delegate individual tasks to AI systems, but someone still needs to decide:
- What the software is supposed to do
- Which requirements are essential
- Whether generated code is correct
- How the system should handle unexpected conditions
- Whether security requirements have been met
- How different AI agents and tools should work together
- Whether the final product is suitable for real-world use
This is why Mesaroš sees the modern developer’s role as extending beyond traditional coding.
The developer becomes an orchestrator of AI-powered development workflows, while remaining responsible for the quality of the resulting software.
AWS Uses Kiro to Address AI Coding Challenges
AWS has also been developing tools aimed at improving the reliability of AI-assisted software development.
One such product is Kiro, an agentic development platform introduced by AWS that uses a spec-driven approach.
The idea is to give AI coding agents clearly defined requirements and guidance before they begin producing software.
Mesaroš said specification-driven development is not a new concept. Developers have used requirements documents and specifications for years to define what software should do and how features should work.
However, creating detailed specifications can be time-consuming.
Kiro is designed to bring that structured approach into an AI-assisted development workflow, with requirements serving as guardrails for the agent.
What Is Spec-Driven Development?
Spec-driven development starts with clearly defined requirements rather than immediately asking an AI system to generate code.
A typical workflow can involve:
- Defining the desired functionality.
- Describing how the feature should work.
- Establishing requirements and constraints.
- Giving the AI agent structured guidance.
- Generating and implementing the code.
- Testing the result against the original requirements.
The approach is intended to reduce ambiguity between what a developer wants and what an AI agent actually builds.
Mesaroš argued that this provides a more structured alternative to simply asking an AI coding tool to produce software from a short prompt.
AWS Highlights Kiro’s Adoption in India
AWS said Kiro has developed a growing user base in India and cited startups including Dhan, Pristyn Care, Analytics Fox and Glimmertech among its users.
The company has also highlighted specific examples of how businesses are using the platform.
According to AWS, Dhan, a stock-trading platform, used Kiro to develop a new real-time charting system supporting more than 170 indicators.
AWS said the project was delivered in eight weeks with a single engineer, compared with an original estimated development timeline of 12–24 months.
These figures are AWS’s stated results and should be viewed as company-reported case-study claims rather than an independently verified benchmark.
Pristyn Care Reports 8% Reduction in Marketing Costs
AWS also cited healthcare startup Pristyn Care as another Kiro user.
According to AWS, Pristyn Care used Kiro for marketing-related functions and reduced the associated cost by 8%.
Again, the figure represents an AWS-reported customer case study and does not necessarily indicate that all companies using AI development platforms will achieve similar savings.
Vibe Coding Has Advantages — and Risks
The discussion around AI-assisted development has also popularised the term “vibe coding.”
Vibe coding generally refers to a workflow where developers describe what they want in natural language and allow AI tools to generate much of the implementation.
Mesaroš said this approach can be useful for rapidly shipping software.
However, he highlighted a potential problem when developers begin introducing sensitive information or building products that require stronger reliability and security controls.
For prototypes and experimentation, rapid AI-generated development can be useful. But production software often requires a much more disciplined process involving requirements, testing, security reviews and human oversight.
Kiro Is Not Designed as Simple Vibe Coding
Mesaroš said Kiro’s approach differs from a typical vibe-coding session.
Rather than simply generating code from an informal instruction, the platform is designed to work from defined requirements.
The distinction reflects a broader debate within software engineering: whether AI coding should focus primarily on increasing the speed of code generation or on improving the entire software-development process.
For enterprise and production applications, the latter can be particularly important because errors can have consequences beyond a failed experiment.
Humans Still Make the Final Decisions
Despite the rapid improvement of AI development agents, Mesaroš does not see human developers disappearing from the process.
Instead, he expects developers to work with AI systems as productivity tools while retaining responsibility for decisions and outcomes.
Ultimately, humans remain responsible for determining what should be built, what constraints matter and whether the final software meets the intended requirements.
As Mesaroš put it, “A developer will sit there and produce software with an engine.”
That reflects the broader shift taking place in software engineering: AI may increasingly become the engine that produces code, but developers remain responsible for directing that engine and determining whether what it produces is good enough to ship.
The Bigger Shift in Software Development
The rise of coding agents is unlikely to make technical expertise irrelevant. Instead, it changes where that expertise is applied.
Developers may spend less time manually writing routine code and more time on:
- Software architecture
- Requirements engineering
- Testing and validation
- Security
- Code review
- System reliability
- AI-agent orchestration
- Production monitoring
- Managing unexpected behaviour
The central challenge is therefore not simply how quickly AI can produce code, but whether developers can build reliable processes around it.
For companies adopting AI coding agents, the ability to combine automation with strong human oversight could become just as important as the underlying capabilities of the AI models themselves.
