Anthropic’s report calls Claude Mythos Preview its most capable AI model, able to outperform nearly all human experts at finding and exploiting software weaknesses. It is not available for general use, yet a risk update says it is already used heavily inside the company for coding, generating data and other internal work. So the strongest models start changing how work gets done inside AI companies before the wider market can get them. For most firms, the limit is the human process around the work.
Internal use, process design and speed of adoption decide who builds an advantage first. As models get stronger, more computer-based work becomes easier to speed up. A study covered 4,867 software developers in three real-world experiments. It found that those using an AI coding tool completed 26.08 per cent more tasks than those without it. A gain that size can affect delivery speed, staffing assumptions and how much a team can produce without adding headcount.
Stronger tools do not automatically redesign an organisation, though. A separate paper studied 7,137 knowledge workers at 66 firms and found that those given an AI tool spent two fewer hours a week on email. Yet the researchers saw no sign that giving the tool to individuals changed how many tasks people did, or what kind. The gains appeared first in work people could change on their own, and not in work that depends on coordination across teams.
Producing the first draft, the first analysis or the first block of code is no longer the slowest part of work. The gains from a stronger model depend on the organisation using it. That organisation has to give the model the right information, review its output quickly and turn faster drafts into finished work without new delays. So a company that waits for a model like Mythos to reach public release before adapting is focused on the wrong moment.
In most firms, the bottleneck is already there: unclear ownership, slow approvals, scattered information, repeated handoffs and review cycles built for a slower pace. As AI models improve, those frictions become more visible and more expensive.
