By Graham Dean, Chief AI Officer at RealityMine

Artificial intelligence is transforming how software gets made, not just for developers, but for everyone. I have explored what AI coding agents like GitHub Copilot, Claude Code, or Kiro are making possible, and what it might mean for the future of software engineering.

From “Pride Coding” to “Vibe Coding”

Traditionally, software was written line-by-line — a craft that required deep technical knowledge and a fair amount of patience. I’ve started calling this pride coding, when developers take ownership of each line, refining and refactoring as they go. But today, AI tools allow us to move further up the abstraction ladder.

At the other end of the scale is what Andrej Karpathy dubbed vibe coding, where the user “chats” with an AI agent, describing what they want in plain language. The AI handles the structure, syntax, and logic, with the human in the role of director or orchestrator.

“Guide Coding”: The New Middle Ground

Between those two extremes lies a new mode I’ve started calling guide coding. It still uses AI, but with more guardrails, prompting the model with clear specifications, coding standards, and integration instructions. You’re still shaping the system, but now you’re architecting, not just executing. Tools like AWS’s new Kiro are designed to support this kind of structured collaboration.

At RealityMine, we are using AI coding agents all along this scale, and learning what works best for each use case.

Why This Matters

This shift isn’t just about productivity, although that’s a big part of it. It’s about rethinking what it means to create software. A few key implications:

• The skillset is shifting. Knowing how to write elegant code may still be valuable, but more important is knowing how to define a problem, articulate a solution, and test outcomes. This may have always been the case for senior professionals, but all software engineers must now learn to fluidly shift along the scale from pride coding to vibe coding and select the most appropriate option for a given task and time.
• Custom software is becoming more accessible. Just as desktop publishing empowered creators in the ‘80s, AI agents are lowering the bar for non-developers to create tools that solve real problems — without needing a CS degree.
• The idea of “who builds software” is expanding. From schoolchildren to small NGOs, more people can now create and customise their own tech — whether for learning, community impact, or just fun.

Yes, There Are Risks

Of course, there are challenges too:

• Hallucinated or inefficient code. AI models can (and do) make mistakes, repeat code unnecessarily, or miss context. That’s why human oversight and testing remain essential.
• Security and compliance. As software creation becomes more widespread, so does the risk of malicious or poorly governed code.
• Ethics and sustainability. From model alignment to energy usage, we’re still grappling with how to make AI development both responsible and efficient — but the tools are already helping improve themselves.

What’s Next?

In the near future, we may see:

• Highly personalised apps created and continuously updated on-demand via natural language prompts.
• Shorter development cycles, driven by rapid prototyping and real-time iteration.
• Shifts in education, where “learn to code” becomes “code to learn.”

And perhaps most exciting, we’ll see software born in more places than ever. Grassroots projects solving hyper-local problems. Back to the days of kids creating businesses from their bedrooms. And professionals like us focusing more on what we build, and less on how we type it out.

The future of software isn’t just faster. It’s wider, more inclusive, and more imaginative.

About the Author

Graham Dean is Chief AI Officer at RealityMine, where he leads the responsible adoption of AI across products, processes, and teams. He joined as the company’s first engineer in 2012 and served as CTO for seven years before shifting focus to AI strategy, ethics, and governance.