The agency scene is shifting fast, and I’m right in the middle of it.
My name is Pierre Gronberg and I’m the founder of Helsingborg Design LAB, now known as HDL.

My background is in business development. My skill is taking ideas to market, products to revenue, and right now I’m doing that with three distinct products: HDL Commerce, Kepler Cloud and what we today call HubForce.

With this post I’m starting something new. A blog series called The day AI killed agency work. I’ll share my thoughts, actions, and how we’re pivoting in this avalanche era that’s hitting us all.

I mean what I say. The raw and bitter truths of my every day. How AI is changing development. How it’s both an existential threat to my foundation and business. And how, strangely enough, I love it. It intimidates me, but it also drives me. The challenge we’re facing is massive, but so is the opportunity.

It all really started three days ago from writing this post. Yes, only three days ago. You can ask where I’ve been this whole time, but the truth is I’ve been here the entire time, building and watching.

I used Lovable before it became mainstream. Today everyone is vibe coding and let’s be honest, it’s amazing.

I’ve used ChatGPT since the early days when it was crashing every five minutes and people thought their careers were over. Now, every email or message I get from 85% of our customers has been through ChatGPT before it reaches me. I’ll be honest, 50% of my own replies are written with AI. It makes my communication sharper and saves me hours.

But let’s not get derailed.

We all know what Lovable is.
We all know what ChatGPT is.
We all know what it does.

The truth now is that you can almost do everything people used to ask me for a quote for.

Let me put that in context.

I needed to gather all our subscription data into one place to get metrics. We have just over 100 active paying customers, ranging from 99 SEK to 190,000 SEK per month in recurring revenue. Normally this takes hours to pull into a spreadsheet, but I only had four hours before a meeting.

So I opened Lovable, started prompting, and 27 minutes later I had a working subscription dashboard. Two years ago this would have been a 150-200 hour project. We would have scoped it, designed it, built APIs, QA’d it, and written “OUT OF SCOPE” on every little change the client asked for after launch.

Hey, 200 hours are nice. But those requests stopped coming almost overnight.

Why?

Developers keep saying AI can’t do what we can do. But in the last 48 hours, I’ve seen it do it better.

Still, there’s a big but. There are things AI can’t do. Just like there are things humans can’t. It all comes down to a few key things:

– The context behind the prompts
– The quality of the scope
– The ability to understand what’s actually needed

I’ve seen both sides. Wrong context, bad prompts, poor scope. The project fails. The cost? In Lovable, it’s time. In real life, it’s money and reputation.

So back to my week #42 project. I decided to review the apps we pay for and see if I could rebuild them internally as SaaS tools. Same value, better fit.

I started with a feedback tracking tool. Something we use to collect bug reports and feedback. It saves hours and makes it easier for our clients to report issues.

So I made it my challenge to rebuild it in-house.

Goals:

– Build a feedback tracking tool with snippets and scripts
– Add ClickUp integration
– Multitenancy and user management
– Stripe payment integration
– Subscription and credit usage
– Multiple implementation methods for different app types

I started in Lovable. No scaffolding. Needed to pay for a Supabase backend. It got messy fast. So I switched to trysolid.com. That’s where things exploded.

Compared to Lovable, Solid was next level. It actually built backends and databases that felt natural to work with, like chatting with my dev team.

First prompt: “Build me a feedback tracking app.”

Five minutes later it was building. Ten minutes later I had a fully functional SaaS app with database, backend, widget, demo pages. Out of the box.

Then came the refining. Every prompt took 5–10 minutes, but that forced me to think clearly. My prompts got better, my planning improved, and I started doing market research in parallel using ChatGPT.

At the same time I jumped into Framer, built a quick landing page, copied a template, changed colors and logo, and wrote some copy. Within an hour I had a product, a brand, and a domain name: NordBug.

Perspective check:
We’ve charged clients 250–300 hours for similar projects. Here I was doing it in an hour.

But to be fair, those 250 hours result in a stable, production-ready product. Mine was still a prototype. I hadn’t reviewed the code or pushed anything live. But still, the gap was shocking.

I kept going. Added unit testing, wrote automation for functionality checks, then moved to multitenancy and user auth. Originally it was public, so I built login and register. Then I tried OAuth with Google. Twenty-five minutes later, I had a working Google login flow.

03:20 CET, laptop closed. 3 hours and 35 minutes of building.

Next day, laptop open, 2 cans of Celsius, leftover tacos. Mission: ClickUp integration.

This was harder. At first I needed an API key, but I changed it to a full OAuth flow so any user could connect ClickUp directly. I had to think about data isolation and integrity. A few hours later,fully functional, tested, verified.

And that’s where I am now.

So what’s the point?

AI can build faster than I can quote a client. But it still needs vision, logic, and context.
It doesn’t know why to build something,only how.

That’s where the human part comes in.

Maybe this isn’t the end of agency work. Maybe it’s just the rebirth of it.
The agencies that survive won’t be the biggest ones, they’ll be the smartest ones.

This is what we’re exploring at HDL.
Real projects, real tools, real change.

The agency scene is shifting fast, and I’m right in the middle of it.
My name is Pierre Gronberg and I’m the founder of Helsingborg Design LAB, now known as HDL.

My background is in business development. My skill is taking ideas to market, products to revenue, and right now I’m doing that with three distinct products: HDL Commerce, Kepler Cloud and what we today call HubForce.

With this post I’m starting something new. A blog series called The day AI killed agency work. I’ll share my thoughts, actions, and how we’re pivoting in this avalanche era that’s hitting us all.

I mean what I say. The raw and bitter truths of my every day. How AI is changing development. How it’s both an existential threat to my foundation and business. And how, strangely enough, I love it. It intimidates me, but it also drives me. The challenge we’re facing is massive, but so is the opportunity.

It all really started three days ago from writing this post. Yes, only three days ago. You can ask where I’ve been this whole time, but the truth is I’ve been here the entire time, building and watching.

I used Lovable before it became mainstream. Today everyone is vibe coding and let’s be honest, it’s amazing.

I’ve used ChatGPT since the early days when it was crashing every five minutes and people thought their careers were over. Now, every email or message I get from 85% of our customers has been through ChatGPT before it reaches me. I’ll be honest, 50% of my own replies are written with AI. It makes my communication sharper and saves me hours.

But let’s not get derailed.

We all know what Lovable is.
We all know what ChatGPT is.
We all know what it does.

The truth now is that you can almost do everything people used to ask me for a quote for.

Let me put that in context.

I needed to gather all our subscription data into one place to get metrics. We have just over 100 active paying customers, ranging from 99 SEK to 190,000 SEK per month in recurring revenue. Normally this takes hours to pull into a spreadsheet, but I only had four hours before a meeting.

So I opened Lovable, started prompting, and 27 minutes later I had a working subscription dashboard. Two years ago this would have been a 150-200 hour project. We would have scoped it, designed it, built APIs, QA’d it, and written “OUT OF SCOPE” on every little change the client asked for after launch.

Hey, 200 hours are nice. But those requests stopped coming almost overnight.

Why?

Developers keep saying AI can’t do what we can do. But in the last 48 hours, I’ve seen it do it better.

Still, there’s a big but. There are things AI can’t do. Just like there are things humans can’t. It all comes down to a few key things:

– The context behind the prompts
– The quality of the scope
– The ability to understand what’s actually needed

I’ve seen both sides. Wrong context, bad prompts, poor scope. The project fails. The cost? In Lovable, it’s time. In real life, it’s money and reputation.

So back to my week #42 project. I decided to review the apps we pay for and see if I could rebuild them internally as SaaS tools. Same value, better fit.

I started with a feedback tracking tool. Something we use to collect bug reports and feedback. It saves hours and makes it easier for our clients to report issues.

So I made it my challenge to rebuild it in-house.

Goals:

– Build a feedback tracking tool with snippets and scripts
– Add ClickUp integration
– Multitenancy and user management
– Stripe payment integration
– Subscription and credit usage
– Multiple implementation methods for different app types

I started in Lovable. No scaffolding. Needed to pay for a Supabase backend. It got messy fast. So I switched to trysolid.com. That’s where things exploded.

Compared to Lovable, Solid was next level. It actually built backends and databases that felt natural to work with, like chatting with my dev team.

First prompt: “Build me a feedback tracking app.”

Five minutes later it was building. Ten minutes later I had a fully functional SaaS app with database, backend, widget, demo pages. Out of the box.

Then came the refining. Every prompt took 5–10 minutes, but that forced me to think clearly. My prompts got better, my planning improved, and I started doing market research in parallel using ChatGPT.

At the same time I jumped into Framer, built a quick landing page, copied a template, changed colors and logo, and wrote some copy. Within an hour I had a product, a brand, and a domain name: NordBug.

Perspective check:
We’ve charged clients 250–300 hours for similar projects. Here I was doing it in an hour.

But to be fair, those 250 hours result in a stable, production-ready product. Mine was still a prototype. I hadn’t reviewed the code or pushed anything live. But still, the gap was shocking.

I kept going. Added unit testing, wrote automation for functionality checks, then moved to multitenancy and user auth. Originally it was public, so I built login and register. Then I tried OAuth with Google. Twenty-five minutes later, I had a working Google login flow.

03:20 CET, laptop closed. 3 hours and 35 minutes of building.

Next day, laptop open, 2 cans of Celsius, leftover tacos. Mission: ClickUp integration.

This was harder. At first I needed an API key, but I changed it to a full OAuth flow so any user could connect ClickUp directly. I had to think about data isolation and integrity. A few hours later,fully functional, tested, verified.

And that’s where I am now.

So what’s the point?

AI can build faster than I can quote a client. But it still needs vision, logic, and context.
It doesn’t know why to build something,only how.

That’s where the human part comes in.

Maybe this isn’t the end of agency work. Maybe it’s just the rebirth of it.
The agencies that survive won’t be the biggest ones, they’ll be the smartest ones.

This is what we’re exploring at HDL.
Real projects, real tools, real change.