So, about a year ago, I took the plunge into AI-assisted coding. At first, I wasn’t even sure what to call it. Was I “vibe coding” before Andrej Karpathy made it a thing? Possibly. More likely, I was fumbling my way around with large language models, trying to see if they could help me build something useful without blowing up.

A year on, I’ve got a few scars, some modest wins, and – crucially – a handful of AI-assisted coding tips for anyone thinking about letting an AI loose in their dev environment. These aren’t coming from an AI researcher or a Silicon Valley wunderkind. They’re coming from someone with 40+ years in software who’s rusty but curious, armed with too much coffee and a passion for getting things done.

Here are my AI-assisted coding tips

1. Think like a teacher: break it down, step by step

In the early days, I’d ask the AI to “build me a web-based back end for some content” and get frustrated when it produced a half-baked mess. Turns out, LLMs aren’t mind readers; they’re more like very clever interns. You have to hold their hand at first.

Now I break things into atomic steps. Not “build the app,” but “create a project folder with these files” or “write a simple API endpoint.” The more explicit I am, the better the results. I even find myself narrating out loud sometimes, like I’m explaining it to a junior dev sitting next to me. (Pro tip: don’t do this if you work in a shared space unless you like funny looks.)

It reminds me of when I first taught myself GUIs in the 90s. Back then, if you skipped a step, your code didn’t compile. The same discipline applies here, even if the AI feels like magic.

2. Raw chat prompts trump fancy agentic tools

I tried some of the buzzy agentic tools – you know, the ones that promise to plan your whole project and write the code while you sip a latte. Sounds great in theory. In practice? They often went down rabbit holes or got stuck in loops of their own making.

Plain old chat prompts in a browser tab turned out to be more reliable. I type, the AI responds, and we work it out. Sometimes I look things up for Claude on Stack Overflow or feed it some documentation from GitHub. It’s not sexy, but it works. Maybe, one day, the agents will get good enough to be trusted with the keys to the kingdom. Right now, I prefer being in the loop.

3. Design first. Code later.

Seriously. This one is embarrassing because I should’ve known better. Early on, I’d dive feet first into coding with the AI, thinking, “We’ll figure it out as we go.” That way lies darkness and despair.

Now, I start by asking the AI to help me design the system. We talk through the architecture, data models, endpoints, and even UI wireframes. It’s like having an imaginary friend that talks back and suggests improvements. You’ve all got those imaginary friends, right? Just me, then. Once the design feels solid, coding flows much smoother.

This approach saved me on a recent project where I nearly painted myself into a corner with a horror show of React dependencies. Lesson learned.

4. Mix and match your AIs

No single AI is good at everything. Claude Sonnet writes beautiful, clean code. GPT-4 is great at brainstorming and debugging. Even Gemini has its moments.

For a while, I treated my chosen AI like a Swiss Army knife. Now I treat them like specialists in a team. One helps me map out the problem, another writes the first draft of code, and a third sanity-checks it. It feels a bit like the old days when I’d bounce ideas around with different engineers to get the best of everyone’s strengths while learning all the time.

5. Don’t be afraid to throw it away

You know that feeling when your codebase turns into a spaghetti mess and you keep patching holes? Yeah, AI does that too.

When it all goes to hell, I don’t waste time untangling it anymore. I chuck the whole thing and start fresh. This is where AI shines – spinning up a new project doesn’t take hours anymore. You can fail fast and iterate. It’s oddly liberating.

Too often in “traditional” software development, I’ve seen engineers “hoard their chips” and cling to a broken codebase simply because of the time they’ve invested in it.

I’ve probably started and abandoned more projects in the past year than I finished in the 2000s. And that’s okay.

6. Recognise rabbit holes early

Related to #5: Sometimes the AI will latch onto a bad idea and refuse to let go. You ask it to fix a bug, and it keeps proposing variations of the same broken solution.

When that happens, I stop, rewind, and reframe the problem. Or I start a new chat. It’s like dealing with a stubborn developer – sometimes you just need a fresh perspective.

7. Small edits only; control the merge

This one was a game changer for me. Once you’ve got something that works, don’t let the AI rewrite whole files. It’s too easy for it to break stuff you didn’t intend.

Instead, ask for small changes and paste them in manually. Or ask for a step-by-step plan before touching anything. It’s a bit more effort, but it keeps you in control.

Part of my rationale for going the AI route was to help me learn coding in areas I’ve not worked in before (e.g. mobile apps). So I always ask the AI to explain why it’s taken a certain approach, and also tell it to ask me if it isn’t sure what I want. I’ve already learned a ton. The irony, perhaps, is that I’ve learned a lot about writing apps while realising that I’ll probably never need to.

8. Know your destination

Are you building an MVP to show investors? Or a production-grade, secure system? The former: AI can get you there surprisingly fast. The latter: you’ll still need human experts to harden and polish it.

Think of AI as a power tool. Great for rough cuts and prototypes. You wouldn’t use it to install a pacemaker.

9. It’s not for everyone

I love tinkering, and I’ve got enough software experience to spot when the AI is talking nonsense. If you’re a non-coder hoping for a magic solution, you might get frustrated fast.

That said, if you’ve got even rusty coding tekkers (like me), AI can feel like having superpowers. Just be honest about where you are on the learning curve.

10. It’s only going to get better

The tools we have today already feel like science fiction compared to five years ago. Give it another year or two, and who knows what’s possible?

I’m not saying the AI will replace developers. But I do think the nature of coding is shifting. The best developers in the future might not be the ones who can write perfect syntax, but the ones who can orchestrate AIs to build things faster, safer, and smarter.

Some Closing Thoughts

So those are my top 10 AI-assisted coding tips. For founders, this is a potential game-changer. In the UK right now, getting even a modest app built by an agency can run £100k+. With AI, a single founder with a bit of grit can get to MVP for a fraction of that.

Sure, no-code tools promised the same thing, but they’re a bit like prefab houses – fine until you need to add an extension or move a wall. AI-assisted coding feels closer to traditional development but faster.

This isn’t to say it’s all rosy. You still need judgment, patience, and a willingness to throw away bad code. But if you embrace the process, there’s never been a better time to build stuff.

So, that’s where I’ve got to. I’m still experimenting, still failing fast, and still amazed by what’s possible.

What about you? Have you tried coding with an AI yet? What’s worked for you?

Chris Mason MBA

Advisor – Leadership, Innovation & Strategy

Chris Mason is a seasoned leader with over 40 years in the tech industry, recognized for his expertise in strategy, innovation, and team empowerment. As the former General Manager of an engineering software house, he guided the company and its teams through the peaks and troughs. And helped create some of the best vibration and acoustics analysis solutions in the world through strategic innovation and an inclusive approach to leadership.

Chris holds an Executive MBA from the University of Winchester, where his research on “Barriers to Open Innovation for Technology SMEs” reflected his passion for fostering collaboration and driving change. Beyond corporate roles, he is a trusted strategic advisor and non-executive director, helping startups and SMEs unlock potential through leadership development, digital/AI adoption and business strategy. His goal is to transform ideas into impactful realities, empowering teams and businesses to thrive in competitive landscapes.