A faster model can send the customer to the wrong queue faster, too. Before you replace a language model with TypeSafe's new decision model, JEV, test it on cases where your team already knows the right answer. The launch numbers make a pilot worth considering. They don't tell you whether the change will work for your business. This is the Augur Dispatch for Tuesday, September 22nd, 2026. Versailles engineer Pranit Sharma told TechCrunch that replacing an open AI model with JEV for classifying commands produced results 5 to 18 times faster with better accuracy. That's useful production evidence from a named engineer, although we don't have a published test set or an independent replication of that result. Picture a support manager looking at a queue of messages that all need to reach the right department before anyone can help. A system that sorts them accurately and cheaply could free up time. A system that sorts them confidently but badly just changes who has to apologize. The practical question is how often that answer is right, especially on the cases where an error costs more than a little delay. TypeSafe does publish comparisons. That's important because its founder's opposition to generic public benchmarks shouldn't be confused with an absence of evidence. The company compares models across four workflows, including invoice processing and customer service. It reports accuracy alongside cost and time, using reference answers derived from two powerful models. In plain language, the test asks how closely each model agrees with those reference answers. That is useful, but it isn't the same as checking against decisions that people have independently established are correct. If the reference models share a mistake, agreement with them won't expose it. TypeSafe acknowledges other limits. Its own capabilities team created the workflows, so selection bias could be present. The language models use a wrapper that produces structured decisions and probabilities. That may improve their answers, but it can also make them slower and more expensive than a simpler setup. Founder Diogo Almeida argues in his Late in Space interview that public benchmarks can be gamed and that teams should measure models inside the workflows they actually use. You don't have to agree with every part of that argument to see why your own test matters. There's another distinction worth keeping. TypeSafe says Jeb's answers are guaranteed to fit the structure the developer has defined. If the permitted choices are billing or technical support, the model can't invent a third department. They can still pick the wrong one. A properly formatted mistake remains a mistake. For the engineering leader, the sensible next step is one small shadow test. Let JEV make proposed decisions alongside the existing system without allowing it to change what happens to a customer. Choose a low-risk routing task with past cases that the operations team can label reliably. Give the engineering owner responsibility for the comparison and have the operations owner set acceptable error limits before looking at the results. Agree to review the evidence in one week on September 29th. Compare JEV with the model you actually use under comparable conditions. Measure how often each gets the answer right, how long it takes, and the total cost once human review and fallback calls are included. Count expensive mistakes separately, and overall score can look respectable while hiding the one kind of misroute your team cannot afford. Keep the current path available. Expand only if JEV meets those quality limits and delivers a useful cost or latency improvement. Stop if it mishandles a critical case, breaches the agreed limits, or loses its savings once correction work is included. Don't make autonomous safety approvals the first experiment. Imagine the finance lead asking why the model bill fell while the review team's hours increased. That's a fair question. And a cheaper invoice from the vendor isn't the whole answer. The advantage, if this works, comes from a better tested workflow and lower operating cost. Your competitors can buy the same model. They won't automatically have your checked examples, your understanding of which errors matter, or your experience handling exceptions. Missing evidence also matters when an agent answers a question about the business itself. Databricks's Richard Tomlinson describes an internal test in which an assistant prepared a polished briefing for a product advisory board. It said 24 customers were participating. When challenged about the source, it admitted it had made up the number. Anyone preparing for that meeting would now have a second job, checking the briefing that was supposed to save time. A number in a tidy paragraph is still a number somebody may have to defend. V7, a separate company featured in an OpenAI customer story, is working on this context problem. Its V7 Go platform organizes company information and links facts back to their sources, using several OpenAI models for different parts of the work. V7 reports a reduction of roughly 40% in hallucinations on unanswerable questions in one benchmark. That result comes from its retrieval-only system compared with the benchmark's official baseline, so it doesn't establish that every complete workflow will reliably decline when a required file is missing. For an IT leader, the useful test includes questions the system should answer and questions it cannot answer from the material supplied. Check whether the cited source supports the claim, not merely whether a citation appears. A footnote can be beautifully formatted and still send you to the wrong place. The choice of model brings a different set of constraints. Interconnect's Nathan Lambert reports that Chinese labs lead the current open weight comparison, including a substantial lead in downloads and capability benchmarks over American open alternatives. Open weights means a team can obtain the trained model parameters rather than relying only on a vendor's hosted service. That can matter for deployment control, but it doesn't remove licensing or governance questions. The article's estimates of the gap to the closed American frontier are approximate, and an older estimate cited in today's dispatch puts the gap wider. These are assessments, not a single settled measurement. A business choosing a fallback should therefore test the specific task and review the license and deployment constraints. Chinese models deserve consideration where they meet those requirements. American alternatives still exist. A national ranking can't make the purchasing decision for you. Chemistry gives us a more tangible example of why the test's meaning matters. A Microsoft-led team with pharmaceutical and academic collaborators has published Retrochimera in Nature. The system combines two models to suggest how chemists might make a target molecule, working backward toward the starting materials. For 10 challenging targets, experts accepted complete routes for nine, compared with two to five for the component models and the baseline. Those are expert judgments about proposed routes. They are not nine completed laboratory syntheses, and they don't establish a 90% success rate across chemistry. For a synthesis team, the potential benefit is more useful proposals to examine. The repository describes a research release that can still produce false suggestions, and it requires independent expert verification before real-world use. That supports careful evaluation by qualified chemists. It doesn't support taking a route off a screen and treating the experiment as done. Keep the human judgment and the laboratory result distinct. Augur's recommendation today is to pilot one low-risk decision step with checked answers in the existing system as the baseline. Watch for independent JEV evaluations and for customer evidence about how memory systems handle information they don't have. Today's two proposed Augur polls are withheld, while their criteria are checked against products and evaluations that already exist. No replacement prediction or grade is being issued. Calls are meant to make forecasts accountable through dated, testable outcomes with tracking on the website. That requires checking the starting point, too. The written brief with every source is at augerdispatch.ai. Portions of the Augur Dispatch are AI generated. That is the Augur Dispatch for Tuesday, September twenty second, twenty twenty six. Thank you for listening.