At today’s AI Engineer Code Summit (AIE Code) the spotlight was on “system one” models — non‑generative models that return structured choices, scores, or probabilities. These models tend to be much faster and cheaper than reasoning large language models, which the summit materials referred to as “system two” models.
Diogo Almeida, co‑founder and CEO of TypeSafe AI, delivered one of the keynote appeals for system one approaches. Almeida was the lead creator of Jev, a product that launched in September — less than two months before the conference — and quickly attracted attention from AI engineers.
When Jev debuted in September, Almeida emphasized the core difficulty of centering software on LLMs: reliability. At AIE Code he went further, calling Jev “reliable intelligence” and arguing that it should be used to “just automate everything.” He summarised the practical problem with an anecdote: you may get “Einstein 95% of the time, but Mr. Bean 5% of the time — and if you can't tell which is which, how the fuck are you going to automate anything?”
Integration with code: a practical argument
Almeida’s key technical point was that system one models belong inside application code. “Code is the language of automation,” he said, “therefore you need to integrate with code.” swyx, founder of Latent Space, interpreted this as an instruction to “liberally sprinkle ‘System One’ model calls everywhere INSIDE application code to make smarter software instead of increasingly convoluted forms of ‘concat strings together and hope for the best’ AI engineering.”
Almeida repeatedly used the phrase “smart software.” The implication is that LLM‑driven systems sometimes fail to behave smartly in production, and that fast, structured components like Jev can provide more deterministic building blocks.
How Glean uses Jev
Chinmay Goyal of Glean followed Almeida and presented a more pragmatic use case. He explained that Glean uses system one models alongside LLMs and other classifiers, and that Jev is particularly valuable for its “zero‑shot flexibility.” As the taxonomy of the decision space changes, Goyal said, they don’t have to continually fine‑tune models.
In a short interview after his talk, Goyal told Latent Space that their customers don’t need to know which model is running underneath: “Jev powers a significant number of flows, but the user just needs to have a product that works; that is always our aim with building anything. So we want to hide away all the complexity of whether we are using Jev or we are using a full‑blown LLM.”
Goyal closed by offering practical rules for when to use Jev; those patterns and prototypes aim to accelerate adoption of system one approaches in production.
Jev as a building block: UiPath and Workday examples
Chibi Vikram and Kevin Bernstein from UiPath, a business orchestration and automation platform, showed Jev‑based experiments next. One demo tested Jev’s ability to rank document chunks: they ran a “tournament” where Jev selected the most relevant chunk from small groups, and winners advanced to the next round. The entire process took under a second, even with thousands of Jev calls.
“We realized that Jev is a building block that we can adjust,” Vikram said, noting the need to find strategies where that building block can be applied. Alone it may seem less intelligent, he added, but with appropriate strategies and patterns, even the current Jev model can produce smarter outcomes.
Scott Roberts, Principal Engineer at Workday, closed the ‘system one’ session with prototypes his team built using Jev. In one example the interface began filling in a time‑off card while the user was still typing — a concrete illustration of Almeida’s idea that a fast model call can change what the software does. Roberts framed it as a shift “from predicting what you want to say to predicting what you want to do.”
New ways to build software
A broader takeaway from the session was that Jev‑like models can open surprising new development patterns. One attendee who leads AI at his company said he had used Jev in his spare time to build a site that sorts and ranks fantasy football news.
Almeida closed by comparing the wave of experimentation around system one models to the early internet. “I think that there's going to be an explosion of what software can do, and we will want people to be exploring those boundaries of software. Not necessarily because it's valuable, but because they can, because we have new frontiers to explore,” he said.
The demos and prototypes at AIE Code suggest that system one models — by virtue of speed, structure and cost — may soon become a standard piece of application logic rather than an exotic add‑on.
