Generative AI is making waves in the software development space, transforming the way engineers operate, iterate, and develop code.
We see the ripple effects of this transformation show up in data management, monitoring, and observability as a whole. AI is changing more than just workflows, it’s changing career paths.
Developers must redefine their roles and leverage their unique strengths to bridge knowledge gaps and adapt to a new way of working.
From hallucinations to self-testing codes, AI is learning from its mistakes
Large language models (LLMs) can be used to brainstorm, collate information, and construct code. However, they still make mistakes and hallucinate frequently enough that they cannot be trusted.
Hallucinating involves presenting false information as true or generating a factually accurate answer that is not relevant to the question it was asked. Such failures can complicate the process further for developers.
Today, developers use AI coding assistants in one of two ways. Firstly, they can employ an AI coding agent to author a smaller, specific piece of the solution.
Alternatively, they use it to help structure their own code more effectively by asking questions about properties of algorithms, checking references, language semantics, and brainstorming the design of the solution.
When the technology was in its infancy, the regular occurrence of hallucinations required engineers to spend an excessive amount of time reviewing and checking code.
This turned it into a time suck instead of a time saver. That’s quickly changing: now, AI coding agents build and run tests against the code they authored and correct their own mistakes. As a result, hallucinations are becoming less of an issue.
Is AI slowing down the developers’ process or speeding it up?
Whether AI speeds up or slows down, the coding process depends on the individual circumstances; much of it comes down to the developer’s level of experience and AI literacy. In fact, a recent study revealed that when developers use AI tools, they take 19 per cent longer than when they do not.
Coding agents offer an additional layer of ideation for developers when drafting code. The challenge is that they often get caught in cyclical attempts to fix their own code. This means that they require significantly higher guidance when working on codebases or in atypical contexts, which might tip the productivity scales.
The more specialized the code and task, the more difficult it is to achieve good results without a lot of guidance or human intervention – particularly when revising and reimplementing the code that the AI is struggling with.
This means that the amount of wasteful time spent is determined by the project itself, as well as the engineer’s familiarity with prompting.
Among the failure rates, 60 per cent of flaws are caused by AI tools, including both small and large issues. The latter includes ‘buggy’ code, which may appear okay at first, but after consideration, the developer realizes it needs substantial fixing.
Incorporating AI into DevOps
One of the best use cases for AI agents today is site reliability engineering. By using a model context protocol (MCP) server that integrates with AI coding tools, like Cursor and Claude Code, engineers can easily integrate AI into their daily DevOps workflows.
MCPs make telemetry data available to AI, enabling it to reason on the data and eliminate the need for manual input of information. This improves efficiency and reduces the likelihood of hallucinations.
This allows s site reliability engineers (SREs) to stay in their flow state inside an ‘editor’ or the terminal and quickly assess the health of service level objectives (SLOs), as well as collating logs, observing error and latency distributions of services.
Bringing this context and making telemetry available to AI to reason on has created a step-change in how quickly tasks can be solved every day. The boost in efficiency and speed leads to higher-performing, happier developer teams who can focus on the unique elements of the job.
They can work to solve business and organizational challenges instead of being hindered by the workload. This is only the beginning of the journey. In the future, specific workflows will become fully autonomous, overseen by a human at the center who will be driving the decision-making and investigations.
The value of AI
For developers, generative AI’s primary value today lies in its ability to sketch out an idea and help brainstorm solutions, providing step-by-step instructions and identifying new areas to research.
The guidance it offers can help young developers improve faster, particularly when they lack the mentorship of a more experienced engineer. That said, whether AI saves or wastes time is entirely dependent on the complexity of the task, the engineer’s level of experience, and their ability to prompt the AI well.
As AI tools evolve and engineers adapt to new working methods, the mindset around coding itself is expected to shift. AI is already transforming workflows, but it will not replace humans entirely.
Instead, it will fill in knowledge gaps within developer teams, provide new ideas on how to structure code, including options that might not have been covered in training before, and take care of the menial tasks that bog down engineers daily.
Ultimately, human-assisted AI will be the most powerful AI.
We’ve rated the best no-code platforms.
This article was produced as part of TechRadarPro’s Expert Insights channel where we 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/news/submit-your-story-to-techradar-pro
