At the AI Engineer Code Summit, Diogo Almeida, co-founder and CEO of TypeSafe AI and lead creator of Jev, used his keynote to urge developers to use 'system one' models to automate far more of their software. These are non-generative models that return structured choices, scores or probabilities, and they are typically faster and cheaper than reasoning LLMs. Almeida argued that LLMs are unreliable as the core of software, joking that they deliver 'Einstein 95% of the time, but Mr. Bean 5% of the time.' Jev was released in September and quickly became the most popular episode of the year for Latent Space. Other speakers showed how they are using the approach. Chinmay Goyal of Glean said system one models run alongside LLMs and other classifiers, and that Jev's zero-shot flexibility means the team does not have to fine-tune models again each time a decision taxonomy changes. Glean hides the choice of model from customers, who only need a product that works. UiPath demonstrated a tournament-style ranking of document chunks in which Jev repeatedly picked the most relevant items from small groups, finishing in under a second across thousands of calls. Workday's Scott Roberts showed a prototype that fills in a time-off card while the user is still typing. For Australian businesses and professionals, the practical change is architectural rather than regulatory. The article gives no pricing, benchmarks or data-handling details, so cost and reliability claims should be tested before adoption. Where a system processes personal information, obligations under the Privacy Act 1988 apply whatever model type is used. The most worthwhile step is a small pilot on an existing high-volume classification or ranking task, compared against the current approach. If a business has no such task, this changes nothing for it right now.