As a software engineer who loves the craft and for whom the most fun part of being an engineer is writing code, the last 12 months have been a roller coaster:

• Fearmongering from leaders of AI labs saying that software engineering is done for as a profession.
• Mass layoffs "due to AI."
• Stories of business executives riding the hype train in full "Dunning-Kruger effect" mode.
• The rise of vibecoders making money with software products.

Not to mention Mythos and the doom and gloom that came out of that one. The list goes on.

What once was a highly regarded profession seems to now morph into all other adjacent roles: designers, product managers, and BDRs now "ship production code."

I think it's impossible not to have felt some level of AI-induced anxiety at some point in the last 12 months if you are an engineer. I was there too. I got over it, and I want to offer you my unsolicited perspective.

One of the incredibly cool things about our profession is that it doesn't require formal education to enter. The barrier to entry has been low for a long time. Literally anyone can code. At the risk of being called an 'elitist', there is, however, a difference between a 'coder' and a software engineer. If you've been in this line of work for some time, I don't need to tell you that writing code is only a part of what a software engineer does. Particularly during COVID, the market was flooded with graduates of code academies or online courses that were sold as 'full-stack engineers' after an 8-week bootcamp. This was actually fantastic news for experienced engineers, who became a lot more valuable, and comps went through the roof pretty much everywhere.

I call "huge BS" on all mass layoff announcements citing AI as the reason. That's just the perfect scapegoat fuelling the fearmongering machine. The real reason is historic overhiring and running low on OPM (other people's money), because unless your pitch deck has "AI" in it, it's not as easy to raise money now as it was 5 years ago. And those companies loved to spend. The early 2020s job market for software engineers was an anomaly; this is a correction.

See, a low barrier to entry also means it's easy to be thrown overboard when the ship starts sinking. I am not saying that all engineers who have been laid off are inexperienced or bad at what they do. Far from it. But most of those experienced, talented engineers will find a job quickly. For the others, I'm not quite sure what to say.

The following may be a hot take, but keep in mind that it's coming from someone using AI every day in a professional setting to generate code as well as do other useful things commercially. The code produced by present-day state-of-the-art models from Anthropic and OpenAI is not great, and the rate of improvement from release to release has plateaued. Those models can produce working solutions from vague prompts, but in most cases, they make poor architectural decisions that over just a few iterations snowball into something that can't be maintained… without them. We are witnessing the biggest ever vendor lock-in of our generation.

The goal of AI labs is to make money, and the way they make money is by selling your output tokens. More tokens = more money. More tokens = more code. What does this mean? AI code is overly verbose (not suggesting lower LoC is always better, or that this is a "token conspiracy" btw!), and even with proper expert guidance and detailed speccing of the work to be done, you still have to iterate even on relatively basic things.

The cost of your output tokens is still heavily subsidised. Just look at how Anthropic has changed their pricing structure in the last month. Look at GitHub Copilot too. OpenAI is surely about to follow suit; they're just bleeding money to max out their cash on hand and compute availability advantage over their competitors.

So without good engineers in-house, an AI-generated code release craze, and continuously increasing token prices, how's the software landscape looking? Quality is going down EVERYWHERE. Look at GitHub. Look at Windows. Look at the Apple OS experience. Heck, look at the AI-generated products Anthropic is putting out. The quality of the software products we're using has degraded since those companies embraced an "AI-first" culture. And this will only get worse as the effects snowball over time.

There are lots of thriving solopreneurs who are openly bragging about vibecoding features while they sleep, encouraging their audience to create products because it's so easy to monetise online and you don't need any software skills to do so. What they fail to mention is that the biggest reason they are able to make money online is that they have a distribution channel: their captive audiences of hundreds of thousands of followers. I would love to see them try selling their vibecoded products without any following.

The single greatest source of practical information for software engineers is (or was) Stack Overflow. The models we have today do what they do a lot better thanks to all the valuable information in there, as well as all of the public (and maybe not so public?) human-produced code in the wild. Nowadays we have AI PRs everywhere, and Stack Overflow is dead. Something to ponder: what will the next-gen LLMs be trained on? AI-generated code, maybe? Hmm…

Maybe you think that after all that I am against AI for coding, or AI in general. That is not the case, and the next section will hopefully clarify that.

If you are a software engineer today, which means you know or even master computer science fundamentals, I believe you are and will be safe whatever happens. Keep yourself upskilled, stay strong on fundamentals, and don't let your abilities fizzle out because you're not exercising your muscle memory. You may use AI to do most of the 'code writing', but if you only do that, you will lose fluency and perspective. Stay in touch with your abilities. Just think about what Google Maps has done to us — how many people have lost their internal sense of direction without a screen telling them where to turn?

Next, AI is here to stay. We are witnessing a democratization of software production. I strongly believe this is not going to work well at all for those who are overly enthusiastic about what they can achieve. It's one thing to see something 'working', and another thing to have a quality software product that can reliably service paying customers. The degradation of quality of established products is an opportunity for others to carve into their market share. Smart engineers will seize this opportunity.

However, there is a huge danger. The vendor lock-in I mentioned earlier is a two-pronged issue. On one side, you have SOTA AI labs capturing demand, locking companies and individuals in, and then raising prices. They are so good at locking people in because their models hide away the lack of abilities of those using them, as I said before. So once you can't afford Opus, you'll move to Sonnet and you'll have a terrible experience, leading to you throwing more money at the problem to go back to the higher tier. Whoever is saying AI will get cheaper in the next 5 years should not be trusted. The prices we see today are not the real prices that those companies have to charge to make a profit.

The second side is that of hardware prices. Chip fabs have now diverted their production from consumer hardware to AI-specific data centres. This means that you and I can't get our hands on powerful enough hardware to run LLMs locally. By the way, open-source LLMs are insanely good for their sizes, because they are meant to work with very different constraints. I encourage you to check the Qwen 3.6 and Gemma 4 series models. Anyway, if you can't buy hardware to run your own intelligence, you end up unable to leave the big vendors. The oligopoly is shaping up already.

I strongly feel it is the time to invest in personal hardware that allows you to run your own AI. It won't be as good as what the big labs release. However, if you know your fundamentals and do all the work software engineering involves prior to code writing, local LLMs you can run with 64GB of fast memory will take you most of the way there. And that is today. Open-weight models have, in my opinion, made more progress than big SOTA models in the last few months.

So what have I actually been saying? To TL;DR it:

• You're safe if you know your craft. If you're a software engineer and know your stuff, there will be lots of work coming your way. You're a hot commodity.
• The price trap is real. LLM vendor lock-in is happening, and we're on a countdown to massive cost increases once AI labs reduce subsidies.
• Own your artificial intelligence. Hardware is getting harder to buy as prices skyrocket. At the same time, local LLMs are getting very, very good. You should be in a position to run models locally. Invest in capable hardware if you can.