I've watched too many people burn weeks chasing hype. But three weeks ago I put Fable 5 into Conductor on my Mac, and it broke something in how I work that I don't think goes back together the same way.

Let me back up…

I've been using AI as a coding assistant since Opus 4.6 came out at the end of last year, mostly on a side project I've been tinkering with. Useful. Genuinely helpful when something wasn't working or I needed a prototype fast. Not revolutionary, though.

Then a few weeks ago Anthropic shipped Fable 5, their new frontier model. And it didn't just make me a bit faster. It rearranged how I work as a software engineer.

My podcast co-host Bo Møller was ahead of me on this. He's had multiple agents running in parallel inside Conductor for a while, and every time it came up on SaaSKøbmænd I'd nod along and quietly not do it myself. Partly because the setup he'd built was his own work, not something I could just copy. Partly, if I'm honest, because I didn't trust the models enough. I didn't believe they were good enough to build something I'd actually want to ship.

Three weeks ago I stopped waiting.

Now I open Conductor, hand it five or six tasks (features, refactors, whatever needs building) and it copies the entire codebase once per agent so nothing collides and implements the new code. Some finish in ten minutes. Some take an hour. Depends entirely on how well I explain what I want.

The workflow: I write the first prompt like I'm briefing a developer. I ask for a plan before any code gets touched. We go back and forth on that plan, sometimes two or three rounds, until it's right. Then I say go, and it starts writing.

Because Herodesk has a properly structured codebase (we spent the first two and a half years building the foundation by hand, in Yii2, documented, boring, correct) the agent understands how the pieces connect. It doesn't just write the feature. It wires up the permissions in the frontend. It adds it to our automation engine so customers can build workflows around it. It documents the API endpoint. It writes functional and unit tests for the models, the API and the frontend. Then it spins up its own Docker container, mounts the workstream, links the dependencies, and runs every test itself before calling it done.

I review what comes back. Sometimes I ask for changes. Then it goes into a merge request, I pull it locally, test it by hand, tweak a bit here and there, and then ship it.

In the first week of July I shipped more than 25 features, fixes and improvements. That's not bullshit, and I'm not rounding up. Twenty-five! In one week!! That used to be a month of my life, easily.

Right now I'm building one of the bigger features on our roadmap for this year. Won't say what yet (but keep an eye on my LinkedIn at the beginning of September - big things are coming!). The first version took Fable 5 about fifty minutes. We've run six or seven more rounds on top of that since. It's touched more than 110 files. It's used 93% of the model's one million token context window, on this one project alone. And almost nothing needs fixing afterwards. What I adjust is UX polish, not bugs.

I tried older models at this kind of thing. They weren't good enough.
This is a different category of tool.
This is a different - no, revolutionary - way of working.

For twenty years, having the best product was a real competitive advantage. Better software, better UX, faster shipping. An actual moat you could build and defend. I don't think that's true anymore, or it won't be for much longer. When a fully tested, properly integrated feature takes fifty minutes and one well-written brief, the technology gap between any two companies in the same category is already small. Soon it'll basically be gone. There'll still be a difference in whether the button is green or blue. That's about it.

Which means the real battleground moves. Branding. Positioning. Sales. And, more than any of those, customer service.

It's not a soft, feel-good statement. It's the actual bet I've made for Herodesk. We've decided we're going to be the best in our category at customer service, even though we build AI tooling that automates a huge amount of support ourselves. Our customers already run AI agents that resolve thousands of tickets a month with zero human involvement, and it works well for the simple stuff: order status, shipping, generic product questions. It does not fully replace a human for something like Herodesk itself (yet), where the product is genuinely more technical and the support question is usually "how do I configure this specific automation for my webshop," not "where's my package."

The uncomfortable part is on the team side. What we are building right now with these agents would have taken ten to twenty times as long by hand a few years ago. Realistically, it would have needed a team of twenty or thirty engineers. We don't need that team, and I don't think most software companies will, going forward. If anything, I think we end up hiring more on the commercial and support side than on product, which is a strange sentence for a founder who came up writing code to say out loud.

The job for engineers changes too. Less writing, more reviewing, more quality control, more deciding what's actually good enough to ship. I don't trust any of this blindly, and you shouldn't either if you're doing the same thing.

The bigger discipline problem sits with product. When implementing something takes an afternoon instead of a month, "it'll take too long" stops being a reason to say no. You need an actual opinion about what the product should be, and the willingness to reject good ideas that don't fit it anyway. That's harder than it sounds. I think a lot of product teams are about to find that out the hard way.

I'm still the same guy who waited on this longer than I probably should have. I'm not waiting anymore, and I don't think anyone building software should be either.

Moving our AI-engine from OpenAI to self-hosted servers in Denmark is apparently a very popular topic. My post about it has surpassed 100k views on LinkedIn.

In other news:

We’ve got some big news coming up in September. I’m thinking about doing a “Summer ‘26 release” to cover it all, slightly inspired by how frontrunners like Shopify, Beehiiv and Apple are doing (in concept - on a smaller budget :))

Below are some of my favourite newsletters. So if you like this one, you should consider subscribing to these, too.