If you're doing any AI coding in your

business right now, you need to

understand the difference between vibe

coding and professional AI coding. Today

I want to talk about the line, you know,

that invisible line that separates vibe

coding from professional coding. This

week I had two separate conversations

about this that I want to share with you

because I feel that they really help

drive the point home.

The first was a video call on on we see

of a guy that runs a a commercial refer

business. They basically kit out

schools, hospitals, and offices. And his

biggest pain point was managing

contractors. You know, sending out jobs,

getting quotes back, tracking documents,

and he found that offthe-shelf pieces of

software either didn't work well enough

where he needed four different

subscriptions duct taped together in

order just to achieve the result that he

wanted. So he heard about claude code

and thought could I actually build this

thing with claude code? Now let me

clarify he is not a developer. His

background is in web design. So he knows

Photoshop, he knows Figma, but he's not

an engineer. But he did have a clear

idea of what he wanted and he had the

pain of living dayto-day in his business

with the inefficiencies. So he prompted

his way through it. No, you vibe code it

prompt by prompt. He built this app with

the vibes and a week later he had a

working app. Then I asked him a question

on the call. I said, "What database are

you using?" And he just stared at me

with a look. It's like I asked him for

the meaning of life. He didn't know the

answer. So he logged in to Versel and

after a few minutes he said, "Oh, it's a

Superbase." Now that's pure vibe coding.

He's building an app and has no concern

over the mechanics under the hood, about

the infrastructure, about the

architecture. He just doesn't care. The

AI handles it and is currently working

for what he needs. It's brilliant. is

the the perfect definition of vibe

coding. But let me be clear, not all

coding, AI coding is vibe coding.

There's a line and I want to draw out

this line for you today.

Vibe coding is what this reverb company

guy is doing. You know, you describe

what you want to the eye and the AI

builds it. The AI picks the database.

The AI picks the infrastructure. The AI

picks the architecture. The person

vibing doesn't need to know all of that.

And for a small internal app, that's

okay. But at the other side of the line,

you have professional AI coding. And

that's something entirely different.

Now, what's similar is that the AI still

writes the code, but before a single

line of code is written, three things

need to happen. One, there's a detailed

PD and spec doc that explain what the

app does in the least ambiguous way

possible. It's not a build me a

contractor app and vibing your way

through the build. It's a set of

markdown files so tight that if you were

to give those to another developer or

let's say three developers in isolation,

they will probably end up building the

same app because the documentation is so

tight. Two, a human engineer has to make

the architectural decisions. These

aren't handed off to an agent. They're

handled by a professional who knows the

consequences of not making these choices

correctly now that could happen in 6

months. And three, before the AI writes

anything, it has to confirm what it's

about to build. You feed in all the

docs, but before it actually starts

writing the code, you ask it to confirm

what it's understood and feedback what

it's about to do. It needs to say

something like, "Here's what I've

understood from your request, and here's

what I'm planning on doing, and these

are the steps I'm planning on taking."

Then the engineer has to read through

the plan and confirm that it's correct

before the AI starts writing the code.

You see, the AI is the muscle. The human

is the brain. The AI does the typing and

the human does the engineering. Vibe

coding is fine when you're building

something small where the the

consequences of something going wrong

are disastrous. But professional AI

coding is what you need when real

customers depend on the software when

data security matters when the app has

to scale beyond let's say I don't know

10 users or something. And that's the

line. It's the same AI tools on both

sides of the line. the code may actually

look the same on both lines of this both

sides of the of the line on the surface.

The difference is kind of invisible

until something breaks and then it might

be too late. So, let me show you what AI

coding, professional AI coding actually

looks like in uh our team

at WU. We're in the process of upgrading

our analytics module. We're basically

adding custom dashboards. you know, the

facility to not just display calls data,

but also display data from external

sources. Think of it like a mini BI

tool. And because of what I've learned,

we're approaching this one completely

different to how we've approached our

previous tasks before. First, we've

written a proper PR, not an agile style

story board on a post-it note or

something. A proper detailed PRD that

says exactly what the app does, how it

works, how it needs to integrate, what

it doesn't do. every single edge case.

Second, our lead software architect is

writing a specification that's so tight

that includes things like the data

models, the schemas, the API structure,

how the system behaves in different

states, the error handling, everything.

And third, we will use clawed code to

generate the code. The documentation

will be there, the architecture will be

there. So in theory, claude code can

just generate the code utilizing the

documentation that we provided. There's

also a new idea that I've had that I

want to I want to test out. I'm going to

call it the anti-ambiguity agent.

Basically, we feed in the PRD and the

specs dock into an agent that has a

sequence of markdown files as

instructions that basically say

scrutinize this PRD and this spec sheet

to extract any ambiguity, any possible

where there's choices that you could

make or anything that could need

clarifying. The output should be a spec

that is so unambiguous that Claude code

can just go and generate the code

without any room for doubt. That's my

plan anyway. Get the spec so tight and

then let Claude code write the code. The

test is going to be can we build

productionra software in a way that's

faster or more efficient than if we had

written the code manually. And I think

we can.

The second thing that happened this week

is a call with a dentist. not my

dentist, but a customer of ours who's

asking for an integration between our

phone system and his and his systems.

He's probably one of the only dentists

I've ever met who's actually good with

computers. He's around 35, techsavvy

kind of guy, and he owns two dental

practices in Surrey near London. His

thing is all about business efficiency

through automation with technology. And

he had a similar problem to the

commercial refri guy. So, he went ahead

and built a complete operating system

for dental practices. He built a CRM

with lead tracking built in, a finance

system so it knows the the how much a

patient has spent at the dentist. A

learning management system so that the

dentists can learn how to manage their

practices more efficiently. And all this

in one single platform. Now, he's not

doing this solo. He's hired a developer

with 10 years of experience, a guy

that's fully bought into AI coding. And

between him and this developer, they

wrote all the documentation in advance,

fed it into clawed code. And now they

have a working prototype of the the

platform and they've been at it for 8 to

10 weeks. I think he said this is a

great example of professional AI coding

used to build an MVP that can scale as

now he's planning on launching this

platform as a service to other dental

practices. And the thing is this dentist

isn't the only person doing this. Right

now at my company, we have three open

requests from customers who've either

built their own custom CRM or they're in

the process of building their own custom

CRM from scratch. This was unheard of a

few years ago. And this is happening at

scale inside small businesses behind the

scenes. It's happening everywhere. And

out of those people doing it, the

founders that don't know where the line

is between vibe coding and professional

AI coding, they're going to find out

sooner or later. It could be at the

moment that they have an influx of new

users that's increases the load and

something breaks or maybe the moment

they have to add some new feature to

their vibe coded app and it requires a

massive refactor and it's not viable to

do. This is not a warning sign to stop.

This is just to for you to know what

line, what side of the line you need to

be on and if you're on the wrong side so

you can plan the transition before it

bites you in the ass.

There's a framework that I read a few

months ago from this guy called Dan

Shapiro that really resonated with me.

He describes five levels of AI coding.

Level zero is basically spicy

autocomplete, something like we'd get on

GitHub copilot, for example. Level one

and two, the human is still the main

driver. Level three and four, the AI

writes most of the code and the human is

involved uh to review the code and to

write the specs and the PRD and just to

check everything at the build stage. And

then level five is what he calls the

dark factory where AI writes the

entirety of the code. He's given the

spec. AI is given the spec and and the

PD and it writes all the code, deploys

it, everything. No human in the loop

beyond the p the point of writing the

specs and the PD. I'm not going to go

into every single one in detail, but

just to let you know that most

businesses out there that are either

software companies or that hire

developers or that write code are

probably at level zero or one. Yet, most

vibe coders are probably at level four

already because all they do is they

write the the the prompt, which you can

consider a small P. They submit the

prompt and then they forget about it,

don't look at the code, and they just

look at they look at the end result when

it's ready to be tested. as in that as

in like not unit tests actually test the

thing live but there is a software

company called strong DM who claims to

be at level five complete dark factory

they say that they don't even hire

developers who write code so one of the

specs in their job descriptions is that

we want to hire developers that do not

write code that's their ethos I mean

we're that's extreme we're not there yet

at Wu I want to get there but we can't

be we're four years into a project. We

haven't documented enough at the

beginning. So, we can't have AI write

all our code just yet, but we're working

towards it step by step. The analytics

uh rewrite or the analytics upgrade is

going to be the first step in that

direction.

So, here's where this leaves me. As the

CEO of a 20 person dev team

transitioning towards true professional

AI coding, we're on the journey. Not

arrived yet, but we're learning. We're

adapting. We've got senior engineers who

are drowning in code reviews. We've got

mid-level guys who are still trying to

get their head around AI coding. And I'm

pushing as hard as I can without

breaking the team. Actually, there's

there's a whole conversation to be had

around the fact that developers not

adopting or not wanting to adopt AI

coding. And that's probably a

conversation for another video. But the

thing is, the line between vibe coding

and professional AI coding is invisible

for most people. Now, if you're vibe

coding an internal tool for your

business and it's behind your network

and it's secure, you're good. Keep

going. But if you're vibe coding a SAS

and you don't have the technical

expertise within your team and it's a

SAS that handles customer data or it's

it's a SAS that your customers depend

on, remember the line. Plan the

transition. Bring on the engineering

discipline before it's too late.

I'm running a software company that

existed before AI coding was a thing and

I'm on the journey of transitioning to a

AI professional AI coding a dark factory

and I also love vibe coding things on

the on the side. It's so much fun. On

this channel I'll show you what works

and what doesn't work for us in my

company. If you want to follow that

journey, hit subscribe and head over to

axelmolist.com

to join my newsletter where you'll

receive an email once a week or maybe

every two weeks, once every two weeks

where I talk about what's working and

what's not not working inside my

businesses. And remember, vibe coding is

for prototypes. Professional AI coding

is for production. The line is almost

invisible right now, but I hope now you

can see it more clearly. See you in the

next one.