Something remarkable happened in February of 2026. With the release of Anthropic’s LLM model Opus 2.6 and OpenAI’s ChatGPT 5.3, LLMs became dramatically better at coding – to the point they could start pumping out decent code at a rapid pace with guidance and validation by human developers.
These releases were accompanied by much fanfare about “agentic” coding, which could either mean LLM’s with access to command line tools that gave them the ability to manipulate and interoperate with other applications, or LLMs that could be set up like a team of human developers to run competitively and autonomously.
At the same time, the massive layoffs that had been occurring over the previous 2 years at Microsoft, Meta, and Amazon accelerated. While in previous years this was explained as a correction after a period of over-hiring since 2020 due to low interest rates, in 2026 this was explained as an attempt to free up cash to spend on data centers for LLM model training and LLM inference.
I’m not deep into the principles of AI, myself. I just work as a software developer on AR and VR projects who occasionally plays with GenAI tools like ComfyUI and the various coding tools. But LLMs, since February, have rapidly taken over the world of software development.
And it occurred to me that I need to document this as the transformation is underway. Having spent the past decade in touch interfaces, AR and VR headsets, and depth cameras, I’ve noticed how quickly something new can become old and then commonplace, while we loose a sense of what our lived experience was like before, take your choice: smartphones, turn-by-turn car maps, podcasts, social media, 4Chan, Tumblr, the Musk buyout of Twitter. This can be bad, since if we don’t remember how things were before these technologies came along, we will have trouble seeing how they change the way that we think about the world.
It’s been four months since LLMs really broke out in the field of coding. This led to a massive influx of cash and hype into Anthropic and OpenAI as the prepare for IPOs this year. Generally, the public statements of companies preparing for IPOs about their capabilities and potential are inherently untrustworthy. This isn’t helped by the fact that the way both companies have tended to market their prowess is by emphasizing how potentially dangerous they are for the future of humanity – on the notion that something that can kill all of humanity can also save it?
In June of 2026 there does seem to be a bit of a blowback. This can be seen by the prevalence of articles actually talking about an AI “blowback”. Companies like Uber and Microsoft are cutting back on their API expenditure on tools like Anthropic Claude and OpenAI Codex, after initially encouraging their internal developers to spend what they needed to improve internal software and learn the tools.
I don’t think this is due to any sort of recognition of an incapacity in the AI tools so much as a problem with scaling up its use. A lot of money, for instance, has probably been wasted trying to get non-tech office workers to use these LLMs to manage their emails and time. That sort of enforced AI use has created a lot of resentment and suspicion that by using AI in these ways, you are simply training your replacement. While in coding-focused roles, it has been a problem of acceptance and disseminating knowledge about how to use LLMs effectively to write or co-write code.
Over the past four months, there has also been a lot of hand-wringing about what LLMs can and cannot do. There has been a rise in software developers emphasizing that they are “senior” developers along with a discourse about the qualities of a “senior” developer – about to get along with others (senior developers have people skills, dammit), intuition about how to put code together, knowledge of how to properly “architect” code in a potentially pragmatic manner. These explanations are also usually accompanied by concern over the fate of “junior” developers, and hand-wringing that without junior developers we’ll never have future senior developers.
It feels like a bit of a slight of hand. By showing concern for “junior” developers, developers with jobs get to distance themselves from another group they are basically saying can be replaced by LLMs, which making a case for why their roles should be inviolate. Part of what is fascinating about this is that pre-2026 and going back 20 years, “senior” developers typically described themselves as being 10x solo devs who knew the intricacies of certain software languages or microservices or TDD so well that other devs would simply slow them down. But no longer.
It also feels like an attempt to head off another inevitable question. Why should companies pay senior developers so much when a decent LLM for $200/mo can give junior developers 10x coding kung-fu powers?
Back to the blowback. Microsoft Co-pilot is not transforming the business world and is instead creating a lot of resentment and frustration. Graduating classes over the past month have been boo-ing commencement addresses that give basically good advice about how this generation of students needs to adapt to and master the new technology. The Trump DOD has taken an adversarial position against Anthropic, which insisted that they wouldn’t allow their tech to be used to blindly target civilians, while the Trump administration has reversed its laissez-faire stance on AI and recently forced Anthropic to take down its two newest frontier models, Fable 5 and Mythos.
It is worth pointing out that one of the most sweeping cultural effects of LLMs at the moment is the disruption of the mechanism for posting and applying for jobs. People are using LLMs to rewrite their resumes to fit job postings while HR departments and recruiters are using LLMs to draft their job postings and then evaluate submitted resumes for those jobs. The general effect is overload on all sides. Every job hunter is applying to hundreds of jobs, most of which they are not qualified for, while companies are turning away a lot of people that actually are good fits for open positions because there is no great way to differentiate the slop from the gold. LLMs appear to have made the job market not only inefficient but potentially dysfunctional.
All of which makes the most valuable person in the current job market a junior developer with good LLM coding knowledge with personal connections at a company looking for software developers. Everyone else has good reason to sweat because LLMs really are pretty decent coders, and its hard to know what we are going to do with this knowledge.
