18 Steps to Vibe Coding Nirvana - Part 02

Last time we talked about vibe coding and app development and perhaps the wrong way to think and go about it. This week, we present an 18 step guide (don't take it too seriously but this is what we do).

So let's try again. I can only say how I do it but I can say I've been doing it probably as long as anyone else on the planet:

1. Come to your desk with between 5-35 years of coding experience, app development, software patterns and architecture and a boundless "kid in a candy store" enthusiasm for building exceptional user experiences for people whilst solving some real problem for them that will improve their lives.

2. Did you pass number 1? Good. Next is discuss your idea with either a person or an AI such as ChatGPT. By the way it is only worth having AI to do this if you have at least a Plus subscription. I don't think Pro is worth it for any of them unless you have money that you can burn. I like to discuss my idea with ChatGPT 5.4 thinking mode. I end up with a ginormous prompt covering all aspects of the app and probably the user interface as well although it is practially impossible to just build an interface that is any good at all without real user feedback and many iterations of failure, improvement and eventually success. I know because I used to churn out diagrams of user components and experiences and then walk the user through them over the phone.

3. Generate and finalise your prompt. You will probably have it phrased to build the entire thing but having done this for nearly 2 years now, I am coming down on the side of building the functionality first - almost like a CLI (command line interface) and only then building a UI on top - decoupled from the app. In addition I would say really think about the prompt and look at it like a set of requirements and a high level design. If you can, put in your architectural no compromises here as well. In any case you will have to refactor it every so often - sometimes majorly and sometimes throwing it away and starting again.

4. Put your prompt in: Replit, Lovable, Bolt, ChatGPT, Claude, Gemini...etc

5. Make a coffee and drink it.

6. Try your app out and make changes

7. Move you app into VS Code or Cursor and start building there. For the record, I used Cursor a lot but then it started costing too much when my requests were taking 1 million tokens. I had to keep increasing my limit so I gave up with it. VS Code and Plus GPT 5.4 with Extra high thinking - I never blow any limits even if I work all day and give it super hard things to work on.

8. Create a Github repository and store it there otherwise 100% you will lose it.

Before going onto step 9 you should understand that this is the beginning of development of the app after step 8. You haven't even really started yet. Unless you want to deploy an unmaintainable, insecure and probably horrifyingly bad application? You probably need to knuckle down and do a lot of thinking about how you are going to make your app great. Its going to take a lot of effort. Sure you'll be going at the speed of light but it will still be hard.

ok - onto the next step then....

9. Use VS Code with a Chat extension (At present I am using both OpenAI and Anthropic. The argument against OpenAI right now is compelling but then logically you would probably have to ban most corporates from your life for being corrupt - Amazon, Meta, Woolworths, Coles...etc A kneejerk reaction possibly isn't the best way forward even though a return to the high would be most welcome IMHO). I'm starting to use Claude Code but at the moment in VS Code because setting up the terminal has so far taken more than 5 mins in Windows - ok so I'm impatient. UPDATE: I used Claude code in terminal mode and in VS Code - my thoughts - it is good but I don’t see any reason to change - plus what is “discombobulating”?

10. We haven't even got onto Clawdbot/Moltbot/OpenClaw yet !!! Actually just forget that unless you are a Linux geek. Its pretty amazing if you want to hand over your permissions to your machine. You should hold your breath and cross your fingers in this case. But then some of us are optimists. UPDATE: The install is probably easier now BUT you need to know what you are doing if you this this software.

11. Be prepared for your AI LLM to build something that is less than optimal with giant files 1000s of lines long, no architectural plan - unless of course you are constantly issuing prompts that are architectural in context and nature - which you should be.

12. You get a bug or worse still your LLM enters a kind of loop where it keeps making the same mistake. This happened to me a lot from 2023 to about December 2025 and I had to fix it myself (bummer). What I found was that the fix was often trivial - it shows that these things are not intelligent at all. Amazing though that they can write such complex code and understand context so well now. I find myself fixing the bugs less and less and just inspecting the code myself from time to time so that I have a good understanding of how it all works (this is important).

13. You fix the bug. Next is docker or no docker? I'm preferring docker for most of my applications now. It is just so fast, transferable and easy to work with. I use Gitbash on VS Code - not Powershell or Command (yuck). Gitbash doesn't give me any issues (not true - I had one issue to do with path rewrites or something) and is a proper shell (sorry Powershell but you are crap even though you are powerful).

14. Decide how to deploy your app and do it. Running a VM is normally the way - I prefer to do this myself. I cannot believe the suggestion that coding has been democratized for the masses. I see the posts on LinkedIn - mostly rubbish. It simply isn't true.

15. Plan out your future releases, how you are going to market, advertise, take money. Oh did we talk about security? No? Ah? You better think about OAuth logins as well then.

16. Let's say you have a deployed app getting users. You have just become one of the 1 in 10000 people. Same as before. Supreme effort and I tip my hat to you! I just hope it isn’t a todo list, exercise app or menu planner 😜

17. Now you have to deal with every single user who emails you with feedback, from the ones who say your app is the greatest thing to the ones who say your app sucks - every single one of them and you need to give them a decent answer.

18. Still want to code an app?

Many Views make for...Confusion

There are many views on this. All you have read so far is my opinion - and a most humble one it is. There are far more experienced and genius level developers out there with opinions on this. Some say - programming by hand is the way to go for now. Some have too much inertia to use the vibe. Some are all out on vibe coding and some are half and half. Some even see it as an affront to their masculinity or something (the males anyway). If that is your only or main skill - its a difficult pill to swallow. Myself? I'm like "What? You mean I don't actually have to write another line of code anymore (after 25 years of it)? Show me the way!!!!

What I would say is that if you are going down the route of vibe-coding to produce something that will actually make it to production, you better do all of the development tasks such as unit, integration and system testing, user testing, use version control extensively and CI/CD if possible (not essential). Create good docs for devs, implementers and users; plan, refactor, re-plan. Work out how you are going to support your app - think carefully about security/authentication, logging, backups, scaling - your deployment platform and your ongoing costs. We haven't even talked about marketing (which we aren't going to).

If you think you can just issue a prompt to Repl.it and build an app, well go for it and the best of luck.