Software Development AI AI assistants

Why We Don’t Let Copilots Fly the Plane: Ivan Gekht on AI Coding Assistants

Svetozar Eliseev

26 Sep 2024 — 1 min read

AI coding tools promise velocity, but often deliver noise. In a sobering CIO.com piece, researchers found that many devs aren’t actually gaining much from coding assistants—at least not in the ways that matter most: productivity, correctness, or architectural soundness.

Gehtsoft CEO Ivan Gekht has been vocal about the limits of generative tools in high-stakes development:

“AI can generate syntax. But it can’t reason about risk, architecture, or edge cases under pressure.”

At Gehtsoft, we use AI deliberately—as a surgical tool, not a crutch. Our devs pair AI suggestions with rigorous review, real-world constraints, and battle-tested patterns. Because in regulated, mission-critical domains, correctness isn’t optional—and code that looks right isn’t enough.

Our verdict? Copilots can help—but only if you already know how to land the plane.

Read the full article: Devs Gaining Little — If Anything — From AI Coding Assistants (CIO.com )

Or its summary here.

The Real Management Failure Is Not Lack of Frameworks. It Is Ideology Replacing Operational Judgment.

Over the last two decades, I have worked with organizations across very different environments: regulated finance, SaaS platforms, operationally constrained delivery, large engineering organizations, transformation programs, and Agile implementations at multiple scales. And one pattern appears repeatedly. Organizations stop managing operational reality and start managing management ideology. Sometimes the ideology

Risk Management under Real Velocity: Push Detection Downstream, Limit Blast Radius

I was recently quoted in a TechBullion article on balancing speed and security in business risk management. The core idea was simple: increasing business velocity does not remove the need for risk management. It changes its economics. This is not a technical nuance. It is a decision problem. At higher

AI and the missing first step in learning

What could be the long-term impact of AI on career progression in knowledge work? I saw a post recently asking whether AI will replace junior Scrum Masters. That question itself is not new. But it triggered a realization I had not connected this way before. And it is not very

Code Is No Longer the Bottleneck. Now What?

An advantage that everyone has is not an advantage. Agentic tools handle boilerplate, tests, and even non-trivial features. If your bottleneck was implementation speed, it's gone. Yet Amdahl's Law still applies: the speedup you get from accelerating one part is limited by the parts that don&

Read more

The Real Management Failure Is Not Lack of Frameworks. It Is Ideology Replacing Operational Judgment.

Risk Management under Real Velocity: Push Detection Downstream, Limit Blast Radius

AI and the missing first step in learning

Code Is No Longer the Bottleneck. Now What?

Cookie Settings