Cory Althoff’s latest essay pushes back against the pervasive 'fast takeoff' narrative dominating AI discourse, arguing that software engineers will not be rendered obsolete by superintelligence in the immediate future. Althoff contends that the panic stems from a misunderstanding of current AI capabilities, specifically the unproven nature of recursive self-improvement (RSI). While industry leaders like Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman predict human-surpassing AI within the next few years, Althoff asserts that the burden of proof remains entirely on the labs.
The Myth of Immediate Superintelligence
The core of the replacement argument relies on RSI, a hypothetical state where AI improves itself without human intervention, leading to an exponential acceleration toward superintelligence. Althoff points out that this milestone has never actually occurred. What developers currently observe is AI serving as a better tool for existing workflows, not a self-sustaining improvement loop. The author notes that while AI labs have billions of dollars riding on the belief in imminent RSI, there is no compelling evidence outside of researchers using AI to assist their own jobs.
GPT-6.2 Astra Exposes Judgment Gaps
Real-world performance data supports Althoff’s skepticism. The recent launch of OpenAI’s GPT-6.2 Astra showcased significant benchmark gains but revealed critical operational failures. Early testers reported that the model frequently stalls and loops when handling long, complex tasks. This behavior highlights a fundamental limitation: AI can write code faster, but it lacks the human judgment required to decide what is worth building. The exponential progress in code generation does not translate to exponential progress in architectural decision-making or product vision.
Shrinking Teams, Not Eliminating Roles
Althoff acknowledges that the industry is shifting, with AI already contributing to a reduction in junior hiring and smaller overall engineering teams. However, he distinguishes between team size reduction and professional extinction. The role of the software engineer is evolving from pure implementation to high-level oversight and judgment. As AI handles more of the linear coding tasks, the value of human engineers shifts toward defining problems and validating solutions, a capability that remains distinctly human.
Key Takeaways
- Recursive self-improvement (RSI) remains theoretical with no verified instances of AI improving itself without human intervention. GPT-6.2 Astra’s launch revealed significant reliability issues, including stalling and looping, despite high benchmark scores. AI demonstrates exponential improvement in code generation but only linear progress in judgment and decision-making. The 'fast takeoff' narrative is driven by AI labs with financial incentives to hype capabilities, not necessarily by observed reality.
The Bottom Line
Stop panicking about your job disappearing next year. AI is a powerful tool that will shrink teams, but until recursive self-improvement is proven—not just hyped by labs with billions at stake—human judgment remains the irreplaceable core of software engineering.