Finding My Way
Course Correction
<17.07.25>
Finding My Way
I've been wrestling with a question that's been gnawing at me for months: what the hell am I supposed to do with my career now that AI is eating everything I thought I knew about software development?
This is where I'm at right now. I'm someone who understands the full stack, can architecturally grasp at least 80% of how to deploy a production-ready application on the web, and can passively read and understand TypeScript. BUT, I can't actively write it anymore. Not without help, anyway.
Enter Claude Code. This tool has become my secret weapon, enabling me to create anything on the web with an extreme form of leverage I've never experienced before. It's use has highlighted something: we're probably 6-18 months away from non-technical people being able to create web applications as sophisticated as those built by senior engineers.
The End of Meritocracy in Web Development
I got into web development because I wanted to be a software entrepreneur—someone like Pieter Levels, building online businesses that could scale with almost zero marginal cost. The beauty of web development was that it was inherently meritocratic. You needed a computer, an internet connection, and that was it. Everything else came down to your work ethic and how badly you wanted it.
That world is gone.
Web development is now pay-to-play. If you're not dishing out £180 per month for Claude Code, you're going to be left behind. Your raw intelligence and work ethic only go so far now. The models are getting better and more expensive, and whilst competition will eventually drive prices down, that'll take time. For now, capital is king.
Here's what's particularly fascinating about using Claude Code: I haven't felt tired or got bored once whilst using it. I would routinely get stuck on something in the afternoon when I was coding without AI assistance, get bored, and then feel extremely tired—that specific boredom tiredness. This hasn't happened to me whilst "vibe coding" with AI. What does this tell me? Is it more creative? I'm not getting bored.
But will there be any value left in this if everyone can do it soon?
At the moment, Claude Code needs to be hand-held through the software creation process. I need to direct it step-by-step through development. Give it free reign and it comes up with garbage, but structured, concrete, narrowly scoped requirements can be executed very well. Someone like me can use this as extreme leverage—but for how long?
I truly believe there will be a point when the AI models are so good at coding that you won't even need to look at the codebase, and they'll be better than the best developers at systems architecture. The question isn't if, but when.
The University Detour
In September 2024, I started my Masters in AI at the University of Edinburgh. I did this because I could see how AI was taking over web development and I wanted an insurance policy. In spring 2024, I thought that within the next three years, anyone could create any web application they wanted—and where would that leave me, a mid-level web developer with dreams of being an entrepreneur?
I thought the answer was AI. If AI was eating developer jobs slowly, then surely if I was creating the AI, I'd be the last to go.
I was extremely excited about spatial intelligence and world simulation. LLMs are all the rage now, but world simulation and spatial intelligence will be required to bring AI to the physical world. It's the next frontier of AI, and I wanted to be at the forefront of it.
But I was extremely disappointed to find no one at my university working on this. The few that were working on anything somewhat related either didn't respond to my emails or put me down when I proposed dissertation topics related to world models. A number of lecturers and academics stated that this kind of AI is very expensive and time-consuming to train, and I wouldn't get very far with my dissertation doing this kind of topic.
I was extremely dejected by this. For the first half of my MSc I was really excited about the prospect of working on world simulation, and then by February of this year I felt lost again. I started to realise that AI is slow to train (potentially days or many weeks for large models), hard to debug with slow iterative feedback, and can be very costly to train. None of this is appealing. Yet again, capital would be required to succeed.
In hindsight, I should not have gone back to university. The best learning resources are all online—none of my lectures were as good as the low-cost AI courses on Coursera by Andrew Ng and colleagues. Learning via lectures is extremely boring and slow, the lecture materials are often rubbish and put together haphazardly. Most lecturers and academics care more about their own research than they do helping students.
You can just do things now. Unis are purely accreditation offices now. Fuck that. Show your work, build stuff and put it out there—this is more important than some certificate from a uni.
I spent £17k on the AI course plus the lost opportunity cost of not working for a year. If I could wind back time, I would have tried to work part-time and learn AI on the side myself, or better yet, try to work at an AI company in the world models/spatial intelligence/video generation space doing web dev, then learn about other areas of the business by being proactive and asking questions, slowly transitioning into a more research engineer/ML engineer role.
The Consultancy Trap
Over the last 3-5 months, I was back on the SWE bandwagon, compelled by the world of AI agents and agentic coding. Maybe this could be what I could do? At least there was some overlap with my previous background as a full stack developer.
I interviewed for a company that seemed pretty exciting. I thought they were building an agentic platform that companies could use to build out pipelines of agents to automate mundane tasks within their business. Upon interviewing, I realised they were just operating as a consultancy, essentially getting a load of recent grads to work as forward-deployed SWEs using Claude Code to churn out agentic pipelines for extremely custom business solutions.
It turns out that every business's use case is a little different, and agents aren't capable enough yet to just be given free reign to operate. Custom pipelines have to be built out for each business. This goes against everything I want out of a career. I will not work for a consultancy doing hourly labour. I have to be working on a product or service that can operate when I sleep, where the labour put into making the thing and the product/service's output are disconnected.
Finding My True North
So what does this mean for me now, and what is my plan? I'm trying to figure this out using the principle of regret minimisation, popularised by Jeff Bezos. What is the path that will optimise for the least regret in the future?
For this, I think I have to look at what makes me curious, what excites me when I think about it. What do I want to work on? If I had the choice, if I could magic my way into starting my own company right now or working at an existing one, purely based on what fascinates me and I am most curious about, it would be something still in the space of world simulation/spatial intelligence/video generation.
I just think this is so amazing. The creative potential! I recently watched this video by RunwayML and I was just awestruck. This is absolutely incredible. Imagine the stories people are going to be able to create with this! There really is just limitless potential here.
When I watched this, I thought about my younger brother. He's film mad, loves TV and film, but rightly so didn't pursue a degree in film as it's crowded at the bottom and there's no money in it, so he's doing a business degree instead. It saddens me to think that he might have to spend his life in some boring-arse business job when he could be out there creating novel AI-assisted digital film experiences.
If I could make this possible for people like him, that would be so meaningful. It reminds me how important it is in life for us to pursue both creativity and mastery in everything we do. It's why so many artsy people are so reluctant to do anything else, even though you're subjected to a life of borderline poverty if you're not the best, purely because these aren't "valuable" skills.
I really hope this changes in the future, when AI and autonomous robots take over all of the currently societally "valuable" skills and the only thing left of value will be human creativity and imperfection.
The Path Forward
I think all of this is just a long-winded way of me saying I'm still trying to find my way. I keep taking short detours before finding my way back to the main path, but the main path is becoming clearer.
I know in my heart of hearts that something in the world (4D) simulation/video generation/spatial intelligence space is where I should focus my time. Even if I can't be of much use there now, I should be investing the time into determining how I can be of use to some company that operates in this domain—initially in the realms of web development as this is my skillset right now, but hopefully I'll be able to transition into something closer to the heart of the AI models and deployment.
I need to do everything in my power to get into one of these companies. Why do I think this is a good idea? Well, it solves all of my problems. For me, it solves the meaning and purpose problem by having me work on the technological frontier of AI. It solves the money problem—whether I build a company in this space or, more likely, work for someone else building in this space, I'll be really well paid. I'll get to meet other like-minded people also working on something incredible.
The main path is becoming clearer.