Flexible Automation vs Agentic Augmentation in Coding

Agentic automation (more commonly known ‘vibe coding’) has certainly been booming in popularity, beyond even just the developer space, with Collins Dictionary calling it the word of the year and even Microsoft’s CEO noting that up to 30% (MSFT ) of the company’s code is AI-generated. This approach to coding is undoubtedly driving productivity, but as with any transformative technology, understanding where and how to apply it most effectively is key to maximizing its benefits.

Developers regularly face challenges like scope creep, interrupting coding sessions, and eating into limited time, so the search for efficiency gains through AI is understandable. But developers also need to consider the ‘human in the loop’ philosophy offered by flexible automation. Instead of drawing on automation in each and every process, it focuses on tedious tasks, cementing developers as the decision-makers at each point in the process. This approach supports skill development while ensuring architectural consistency across projects. 

The rise of agentic automation

Vibe coding might be everywhere, but it’s still a relatively new approach, having only been coined in early 2025. It’s the process of using generative AI to produce software code based only on conversational prompts, typically with little to no manual intervention.

It’s been widely praised for lowering the barrier to entry for non-engineers to test ideas and generate working concepts. For example, CEOs and C-suite executives can now demonstrate their desired changes through vibe-coded prototypes, avoiding lengthy conversations with developers where they explain abstract ideas.

But going beyond this ideation stage requires understanding AI’s current capabilities. AI works within certain constraints when handling large context windows, which affects the level of detail in code generation for large-scale professional projects. While developers can instruct it further to make changes if errors are spotted, AI-generated code sometimes duplicates functionality, which can create maintainability considerations. This becomes especially relevant when working with embedded systems that are often constrained by hardware limits, requiring only the tightest code in order to function effectively.

The widespread adoption of AI in coding also raises important questions about skill development. 42% of developers using AI in their processes say at least half of their codebase is AI-generated. As agentic automation becomes more prevalent, it’s worth considering how junior developers build foundational skills. It’s a rite of passage for them to cut their teeth on those routine coding tasks that sharpen their skills and allow them to build coding experience fast. Finding the right balance, where AI handles appropriate tasks while preserving opportunities for hands-on learning, will be crucial for nurturing the next generation of developers.

Developer sentiment also reflects this period of adjustment. In 2024, 70% of developers held positive sentiment towards AI, but this year it dropped to 60%, with 46% expressing concerns about the accuracy of AI code. Yet the majority of developers (70%) do not view it as a threat to their position, and 59% of senior developers in a different survey said that AI tools help them ship code faster. These numbers suggest developers are actively figuring out how to integrate AI effectively rather than rejecting it outright. The technology is changing quickly, and with it, so are best practices for implementation.

So instead of this ‘all or nothing’ approach, it’s worth considering a different philosophy that takes a more measured approach to AI use, keeping developers in the driving seat.

What is flexible automation?

Where agentic automation integrates AI across the development process, flexible automation takes a strategic view.  It counsels for the targeted integration of AI into the coding process, suggesting the replacement of one admin task at a time. This way, the developer always retains control and oversight over the product without excessive disruption. It selectively targets the more repetitive admin tasks, such as code documentation, unit test creation, and any repetitive coding.

Crucially, it recognizes the current capabilities of AI in coding – while it can’t create a full software stack yet, it can drive immediate benefits in some specific areas. So rather than developers becoming frustrated as they apply AI to the wrong tasks, its use is focused on areas where it excels. Over time, developers can familiarize themselves with it and adopt it at a slower pace, allowing its value in solving admin tasks to become clear. Then, developers can return to the more complex, core reasons that they entered the industry in the first place, such as simply writing good, complex software, and solving challenging problems – all while being confident in the AI working alongside them.

Importantly, it also leaves room for a reasonable amount of those routine tasks for junior developers to learn from hands-on experience, allowing them to build up foundational knowledge with the deep learning that comes from traditional trial and error. Rather than being viewed as something that might limit learning opportunities, AI is embedded as a tool – one that developers remain firmly in control of.

The benefits extend beyond individual developers to entire development teams. By automating the repetitive elements of coding, teams can maintain consistency in their documentation and testing practices while freeing up senior developers to mentor junior team members and focus on architectural decisions. This creates a healthier development culture where AI augments human expertise rather than attempting to replace it.

Balancing the automation tightrope in coding

It’s worth reiterating that AI is the single biggest change to coding in decades, and it undoubtedly has the potential to transform how we code for the better, but we need to get that balance right. This is a process that needs to be undertaken strategically as both the industry and developers get used to this new way of working, ensuring we build on strong foundations while embracing innovation. The key is finding the sweet spot where automation improves productivity without compromising the depth of understanding that makes great developers.

This all being said, it doesn’t mean we need to view flexible automation and vibe coding as competing philosophies, but as tools suited for two completely different stages of the software lifecycle. Going forward, vibe coding will be essential for initial ideation, as well as non-technical communication between developers and their wider organizations. And then, as we move into the production process, flexible automation needs to come to the forefront, ensuring that AI remains a help, rather than a hindrance. So, it’s not just one vs the other – to succeed, we need both.