The people who write software may now be the slowest part of building it.
According to Tech Times, an Anthropic engineer who works on Fable 5 argues that in agentic coding — where AI models can take on multi-step programming tasks with limited human intervention — the constraint has shifted. The model is no longer the thing holding progress back. You are.
The framing, as reported by Tech Times, is that the bottleneck in agentic coding is now the developer rather than the model. In other words, the tools have advanced to a point where the limiting factor is how quickly and clearly a human can direct them, review their output, and decide what to build next — not the raw capability of the AI doing the work.
That is a notable reversal. For much of the recent history of AI-assisted programming, the assumption was that models were the weak link: they hallucinated, misunderstood requirements, or couldn't handle large tasks. The claim attributed to the Anthropic engineer suggests that gap has narrowed enough that human judgment, oversight, and speed have become the pacing element.
The source item does not detail the specific evidence, benchmarks, or workflows behind the claim, so those specifics remain unstated here.
Why it matters: if the person guiding the AI has become the bottleneck rather than the AI itself, it reframes what "getting better at coding" means — shifting the emphasis from waiting for smarter models toward improving how developers steer, review, and collaborate with the agents already at hand.