Engineering discipline in software development has been under the spotlight since AI started being used by devs to generate code, with many teams worried it’s causing unnecessary risk. Headlines have been full of cautionary tales about bugs created by hastily shipped code or senior executives taking vibe coding into their own hands.
Nevertheless, teams are finding ways to build trust in AI-generated code. Google’s 2025 DORA survey concluded that AI has an amplifying effect on organizations’ processes – quality processes enhanced by AI lead to higher quality outputs, and more of them.
Our own research shows that most teams aren’t being haphazard with safeguards or accepting AI outputs at face value. Most teams are being more rigorous and giving their outputs the same scrutiny as the work of a teammate. That tells us a lot about how the best-performing teams are learning how to work with AI.
Seventy-six percent of enterprise teams are reviewing AI-generated work at least as rigorously as human-written work. Teams using AI to accelerate their work are seeing results from using proven engineering practices to ensure code is reviewed and tested sufficiently.
Rather than being inherently unstable, the rise of AI-generated code has emphasized that guardrails are the bedrock of consistently reliable software.
Treat AI as a virtual teammate
Teams treating AI-generated code as if it was produced by a human tells us something about how mature teams are using AI. The engineers using AI most effectively treat it as a virtual teammate. They stay in charge of the decisions and they use AI to accelerate the execution.
To make AI adoption an iterative process that improves over time, starting with low-stakes, repeatable tasks and applying stringent checks to outputs gives the tech a chance to work properly without expecting immediate results.
As MIT’s Computer Science and Artificial Intelligence Laboratory reported, AI coding’s widespread adoption doesn’t mean it can handle all aspects of large-scale software engineering by itself. Releasing to production is higher stakes than writing code, for example. Scrutiny from real humans is vital to prevent errors and hallucinations from impacting the business.
At this early stage of AI being used in workflows, it’s positive that 43% of teams are treating AI code like human-written code, with 33% applying even stricter checks. Even the best developers on a team aren’t above the processes and guardrails that guarantee reliable software deployment at speed. AI should be approached in the same way.
Trust in AI follows the risk curve
As with any new technology, introducing AI to the software lifecycle is a learning process. Our data shows trust in AI varies greatly depending on the stage of the software lifecycle, indicating that teams are building confidence with appropriate caution.
A pragmatic approach is being taken at the best-performing enterprises. Thought leaders from Gartner, Forrester and Google’s DORA team all highlight that organizations seeing success from AI-assisted development are strengthening their engineering controls and building up success over time.
This tallies with where AI is trusted to perform. The vast majority (82%) now use AI during the build stage, dropping to 58% at release where production risk is highest. Using AI primarily for earlier stages of software development is a rational step to make sure failures don’t impact the wider business.
This doesn’t betray a lack of confidence in the technology. Almost half (46%) of teams are confident in the performance of AI-generated code, indicating that its more cautious use in production is a measured business decision rather than skepticism.
Outputs are much easier to review and refine before the release stage. With many businesses still lacking full observability, teams often only hear about mistakes in live code once users notify them. Despite claims that software development could eventually be fully automated, teams are showing a clear awareness of where human oversight is essential.
Rather than making blanket judgments about AI’s capabilities, they are taking a more nuanced view, which bodes well for the future of AI-assisted software delivery.
Combine speed with discipline to win
There is continuity in how mature teams are making sure AI-generated code is fit for purpose, but it’s still having a seismic impact on the role of software engineers. When I speak with developers, their feedback is unanimous: the time AI saves is invaluable for focusing on neglected parts of their process.
AI moves the cognitive load from writing code and building configuration to reviewing and directing it. With time freed up, teams have more scope to prioritize observability, test coverage or disaster recovery scenarios – crucial elements of software hygiene that too often get overlooked.
AI adoption is increasing documentation quality according to the DORA report, suggesting that many teams are taking the opportunity to improve the broader engineering practices that support reliable software delivery.
As it takes on more of the mechanical aspects of software development, engineers become increasingly responsible for the work that matters most: validating outputs, understanding risk, and ensuring outputs align with business objectives. If someone asks “why did we build it this way?” it’s a problem if no one can answer the question without asking AI. Humans must be in the loop.
The organizations that benefit most from AI will not be those with more automation, they will be the ones that combine AI-driven speed with engineering discipline and a strict adherence to repeatable processes.
AI adoption is a DevOps challenge
The debate around AI-generated code often focuses on whether the technology can be trusted in DevOps. In most cases, teams have been building this trust iteratively without throwing away human judgment to get the best results.
As AI capabilities continue to improve, the gap between high-performing teams and everyone else is unlikely to be determined by access to the latest model.
The engineers and teams that will thrive are the ones who pair AI’s capabilities with their own judgement.
We list the 10 best vibe coding tools.
This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit
https://cdn.mos.cms.futurecdn.net/PRCsQfoXPXi2t4jsGwWr6L-2560-80.jpg
Source link
