Streamline. Automate. Accelerate. These are just some of the terms often used when discussing artificial intelligence—each reflecting the technology’s core ability to speed operations and reduce manual intervention. Today, AI can generate code within seconds. The technology continues to drive transformative change in software development, and its influence on coding will only continue to deepen.
As a junior developer-turned-CEO with more than a decade in the industry, I’ve seen this evolution firsthand. But AI’s ability to streamline, automate, and accelerate the coding process doesn’t automatically equate to better-quality code. AI solved the bottleneck that was code generation. A once time-consuming and expensive process can now be done faster than ever. When code generation becomes cheap and abundant, the bottleneck is no longer writing the code itself but ensuring the right systems are in place to allow companies to move at AI speed while maintaining the understanding, control, and accountability needed to trust what they ship.
Without proper oversight into the steps that follow code generation—reviews, quality control, and security verification—teams can end up moving faster in the wrong direction. It can lead to slowed delivery cycles, increased maintenance costs, and, ultimately, reduced code reliability. Once code is generated, a human still needs to understand it, review it, test it, and determine whether it belongs in the system. Maintaining that human-in-the-loop approach means there’s accountability for shipping that code. This challenge becomes even more pronounced as more contributors outside traditional software engineering roles enter the development process.
MORE CODE DOESN’T EQUAL MORE PRODUCTIVITY
AI is making it drastically easier to produce software and expand who can actively build it. The AI coding market has exploded over the past year, with a growing roster of new tools—from Cursor and Claude Code to Codex, Windsurf, and dozens of newer entrants—competing for developers’ attention. These tools are equipped to write code, run tests, and identify bugs, among other functions. And a recent Zapier report found that 34% of people shipping software with AI tools have no formal background in programming.
Expanding who can build software is a positive development. However, the challenge is that the ability to produce code can now scale much faster than the expertise required to understand its implications. As a result, organizations need to make sure code generation never becomes disconnected from code understanding. Organizations need stronger systems for deciding what gets approved and shipped, not simply more ways to generate code itself.
There is already evidence that faster code generation does not necessarily mean faster software development and increased productivity. A 2025 study by METR of open-source developers found that when using AI tools, developers take 19% longer to complete issues despite believing these tools had made them 20% faster. This gap highlights a shift in where engineering effort is spent.
AI reduces the cost of code generation, but it does not eliminate the work required to review, validate, secure, and integrate that code into existing systems. When generation outpaces those processes, quality begins to degrade, technical debt accumulates, and reliability suffers.
This is why control becomes the new bottleneck. More AI-generated code means more code that needs to be reviewed and verified. Increasing the number and usage of AI agents means more decisions to govern and audit. More model options mean more choices to manage. The challenge for organizations is not simply increasing the volume of software they can produce, but maintaining control over what gets built and shipped. AI can broaden who is capable of producing code. Governance determines whether that code is fit for production.
So, if control is the new bottleneck, how can CEOs build the infrastructure to manage it without slowing adoption?
TURN CONTROL INTO INFRASTRUCTURE
Organizations need to build control into the way software gets developed. Baseline governance—policy enforcement, usage attribution, and cost visibility—should not be an afterthought or a layer of bureaucracy added after AI adoption has taken off. It should be part of the infrastructure that makes AI adoption safe, scalable, and sustainable from the start.
This serves developers, not just their organizations. With ungoverned and shadow AI lurking in the background, the person with the most to lose may be the individual developer who, without a clear audit trail, established guardrails, or the ability to understand how that code was produced, is held accountable for code they did not fully write. Good governance gives developers the control and visibility they need to work professionally with AI while also giving organizations the confidence to let them move faster.
The goal, then, is not to choose between control and speed. It’s to build the systems that allow both to scale. AI has made code generation faster and more accessible. Organizations that pull ahead will be those that build the governance, expertise, and infrastructure to control what gets built and shipped at the same speed.
Kirill Skrygan is CEO of JetBrains.