Agentic Development vs. Vibe Coding: How Avito Keeps AI-Written Code Under Quality Control

In 2026, Avito published a write-up on Habr about how to keep code quality high when developing with AI agents. Andrey Brovko, head of testing at Avito Auto, noted that on the SWE-bench benchmark, models grew from 2% of tasks solved in 2023 to more than 80% in 2026 over three years, but generation speed alone does not guarantee quality.

How does agentic development differ from vibe coding?

Agentic development differs from vibe coding in how much control it keeps over the process. In vibe coding, the user simply asks the model to "make" something and gets an unpredictable result. Agentic engineering guides the agent through a methodology — specification, tests, review — and makes the result reproducible. In Andrey Brovko's formulation, vibe coding suits informal tasks, but only a structured approach can be scaled across a company.

"AI agents generate code in seconds, but code generation speed does not mean product quality.

If development processes are weak, agents simply deliver defects faster," — Andrey Brovko, head of testing at Avito Auto.

What did the pilot with 100 engineers show?

Avito's pilot with 100 engineers found no statistically significant gain in objective efficiency metrics, and quality metrics stayed at their previous level, even though the team reported high subjective satisfaction. Avito interprets this as the early phase of the DORA J-curve of learning: first people master the tool and learn to verify generated code, and only then does a useful effect arrive.

The payback economics are forgiving. According to the article's calculation:

• SWE-bench: 2% of tasks solved in 2023 → more than 70% in 2025 → more than 80% in 2026
• Pilot — 100 engineers, objective metric gain is statistically insignificant
• Developer salary — $2500/month, AI assistant subscription — $100/month
• A 4% speedup is enough to break even — that is 6.4 hours per month
• Recommended model context window — at least 32K tokens

To support this, Avito deployed three internal services in 2026: skills-hub (a skills marketplace with quality and security checks), mcp-hub (agent access to the codebase, documentation and bug tracker), and AVIDA (Avito Developer Assistant) for safely running agents autonomously, on a schedule or on request.

Where should testing focus?

The focus shifts left — to the specification rather than the final code. Avito builds its process around Spec-Driven Development (SDD). First, a structured specification is written, with user stories and acceptance criteria. Then tests are written following TDD (these also go through human review), and only after that comes the implementation. Each acceptance criterion links to a specific test, which gives 100% requirements coverage by tests. The article mentions SpecKit from GitHub, Superpowers and OpenSpec as ready-made frameworks.

Priority for manual activities shifts to what models still do poorly: reviewing the specification, assessing product and project risks, reviewing tests and code, and exploratory and UX testing. A separate risk is that models can "lie": they report success when only stubs have passed. Against this, Avito uses mutation testing and a combination of human-in-the-loop and agent-in-the-loop, and tracks progress with DORA metrics — Deployment Frequency, Lead Time for Changes, MTTR and Change Failure Rate.

What does this mean?

Avito's conclusion is straightforward: with weak architecture, testing and release processes, AI agents simply speed up the delivery of defects, while with strong ones they deliver speed without losing stability. The company is betting not on the model itself, but on the processes around it.

Frequently asked questions

How does agentic development differ from vibe coding?

Vibe coding is prompting a model with no control over the process and an unpredictable result. Agentic development guides the agent through a methodology (specification → tests → implementation, with review at every step), so the result becomes reproducible and suitable for scaling.

When does an AI assistant pay off for a team?

According to Avito's calculation, with a developer salary of $2500 per month and a subscription of $100 per month, speeding up work by just 4% — about 6.4 hours per month — is enough for the tool to pay for itself.

What is Spec-Driven Development?

It is an approach in which a structured specification and tests (following TDD) are written first, and the code comes last. Each acceptance criterion is tied to a test, which ensures 100% requirements coverage by tests.