How AI Is Changing Custom Software Development

Three years ago, writing software was almost entirely manual work. Today, much of a project's code starts as a conversation with an assistant: GitHub Copilot, Cursor, Claude and others.

For your business, that sounds like instant savings. In part, it is. But the latest data tells a more nuanced story: AI speeds things up, and it also introduces risks you cannot see at a glance.

This article sums up what the research shows and what it means for any company investing in custom software.

AI is already on the team

Adoption is no longer a trend. In the 2025 Stack Overflow Developer Survey, 84% of respondents said they use or plan to use AI tools, and about half of professional developers use them daily ^[1]. Actual use grew from 44% in 2023 to 79% in 2025, and an April 2026 pulse survey found that AI agent use nearly doubled, from 31% to 59% ^[2].

Google's 2025 DORA report, based on nearly 5,000 professionals, reaches a similar conclusion: AI adoption in software development has reached 90%, with a median of two hours of use per day. More than 80% say AI has improved their productivity ^[3].

Faster is not always further

The feeling of speed is strong. The measurements are less clear-cut. A controlled study by METR, published in July 2025, followed 16 experienced developers across 246 real tasks. With AI, they took 19% longer on average, even though they believed they had been 20% faster ^[4]. It is a narrow study using early-2025 tools, but it shows something important: feeling productive is not the same as being productive.

Trust has not kept pace with adoption either. In the 2025 Stack Overflow survey, only 33% trust the accuracy of AI output, while 46% distrust it. The most cited frustration, named by 66%, is solutions that are almost right, but not quite ^[1]. DORA calls this a trust paradox: heavy use, limited trust ^[3].

The hidden risk: security and technical debt

The problem with almost-right code is that it compiles, passes the demo and ships to production. That is where the costs show up.

Veracode tested more than 100 large language models on 80 coding tasks. In 45% of cases, the generated code introduced OWASP Top 10 vulnerabilities. In Java, the security failure rate exceeded 70%, and models failed to defend against cross-site scripting in 86% of cases ^[5]. Most telling: models are getting better at writing code that works, not at writing code that is secure.

Our research reveals GenAI models make the wrong choices nearly half the time, and it's not improving.

Jens Wessling, CTO, Veracode, 2025

In practice, AI-generated code tends to carry these risks:

• Common vulnerabilities: injection flaws, missing validation and unsafe data handling.
• Outdated or unnecessary dependencies that nobody reviewed.
• Duplicated, inconsistent logic that is hard to maintain.
• Missing tests: the code works on the happy path and breaks at the edges.
• Improvised architecture that makes every future change more expensive.

What it means for your custom project

AI can cut time on repetitive tasks, prototypes and documentation. That is real, and worth using. But responsibility for what reaches production is still human.

If your company is hiring a development team, or if your team has already built an app with AI help, ask three questions: who reviewed the code for security, who confirmed the architecture can support growth, and whether automated tests protect what already works.

Using AI is not the risk. The risk is using it without a review process that catches what the model misses.

Next step

At ISSMAR Developers, we use AI where it adds speed and apply expert judgment where it matters. If you already have an app built with AI, our AI Code Audit reviews security, quality and technical debt, and gives you a clear remediation plan. If you are about to start a custom project, let's talk about building it right from day one.