Eldar Azamatov: Testing What AI-Native Engineering Looks Like in Practice

Azamatov is focused on the engineering systems that keep speed from weakening software quality.

One of Eldar Azamatov’s most significant engineering initiatives involved developing and implementing a production workflow at Outtalent in which artificial intelligence generated approximately 94% of the code under his direction. Azamatov designed and supervised the workflow while remaining responsible for architecture, product requirements, technical decisions, documentation, testing, review, deployment, reliability, and security. The result did not convince him that software engineers were becoming unnecessary. It showed how much deliberate engineering is required once generating code becomes one of the easiest parts of building a product.

“People look at how much code the AI produced because that number is easy to measure,” Azamatov says. “I care more about whether the product is reliable, secure, and useful. Someone still has to be responsible for the result.”

Azamatov is a software engineer and technology entrepreneur who previously served as CTO and co-founder of Outtalent after joining the company as its first software engineer. He built several of Outtalent’s core technology platforms, including InterviewDB and OffersDB/Launchpad. He also developed AI-assisted systems supporting education, recruitment, applications, and internal operations. His work gave him a direct view of what happens when a team can generate software faster than it can evaluate and prepare that software for production.

AI coding tools can now examine large codebases, implement features, identify bugs, write tests, and complete assignments that once required repeated work from several engineers. Azamatov believes this shift is changing what companies should expect from technical professionals. Engineers are spending less time producing every line manually and more time defining the problem, organizing the system, reviewing the output, and deciding whether the result is ready for use.

The workflow Azamatov developed at Outtalent was designed to make the codebase easier for both human engineers and AI coding agents to understand. Clear documentation gave the agents structured context about the product and its requirements. Consistent architecture and coding conventions reduced the likelihood of changes that conflicted with other parts of the platform. Defined testing, review, and verification procedures created boundaries for evaluating the code before it reached production.

Azamatov describes this approach as building an agent-ready engineering environment. The AI was not asked to generate code without guidance. It worked inside a system that established how the codebase was organized, what standards changes had to meet, and how errors or inconsistencies would be identified.

“The AI needs context and boundaries,” he says. “Without those, it can produce code that looks convincing but does not fit the product or the rest of the system.”

That structure becomes particularly important when a team moves beyond an early demonstration. Azamatov sees a place for the loosely organized approach often called “vibe coding,” especially when developers are exploring whether an idea has potential. The risk appears when a team treats a fast prototype as though it were ready for customers.

“A prototype can help you learn whether an idea is worth pursuing,” he says. “The mistake is assuming that the first version is ready for real users just because it runs.”

Azamatov’s view is grounded in years of building products rather than observing the shift from a distance. Before joining Outtalent, he helped create and manage consumer games that collectively reached more than 30 million plays and over one million monthly active users. Those projects taught him that technical effort matters only when the final experience gives people a reason to use it. That same principle shaped his work on Outtalent’s platforms and the AI-native workflow supporting them.

His responsibilities expanded after he joined Outtalent as its first software engineer. Azamatov moved from writing software himself to leading product development and engineering processes. He also recruited and managed engineers, communicated with founders and customers, and took responsibility for the business consequences of technical decisions.

The approximately 94% AI-assisted code figure reflected that broader leadership rather than simple use of a coding tool. Azamatov had to determine which work could be assigned to AI agents, what information they needed, and how their output would be reviewed. He remained accountable for whether the completed software met the product’s requirements and could be deployed safely.

This experience has also influenced how Azamatov thinks about developing engineering talent. AI can now generate a functional solution before a newer developer fully understands why it works. He believes engineers must learn how to define problems, evaluate technical choices, and recognize when generated code creates risks that may not be immediately visible.

His advice is to use AI as a multiplier rather than a substitute for understanding. Documentation gives coding agents useful context, while testing exposes errors that fluent-looking code can hide. Engineers also need enough technical knowledge to challenge the tool instead of accepting its first response.

Azamatov shares that perspective through his work as a Cursor Ambassador, educator, mentor, and technology community organizer in Central Asia. He works with developers and technical communities adopting AI-assisted engineering practices. His focus is on helping them understand how to organize codebases, development processes, and team responsibilities around the new tools without weakening software quality.

“Every team can open the same tool,” he says. “What separates them is whether they have the judgment and structure to use it well.”

Azamatov expects small AI-assisted teams to take on projects that once required much larger engineering departments. Their success will depend on more than the amount of code they can generate. They will need experienced people who can define the product, shape the engineering environment, and verify that the completed system is ready for real use.

The 94% figure does not show that engineering has become automatic. It demonstrates what can happen when AI operates inside a deliberately designed production system. For Azamatov, the valuable engineer will be the person who can direct the tools, understand what they produce, and remain accountable for the software after the code is generated.