[KongchangAI]
· 1 min read· 988 words

The Jev Revolution: What Engineers Are Building with System One Models

The Jev Revolution: What Engineers Are Building with System One Models

Jev's creator calls for automating everything as System One models enter developer toolchains and enterprise products.

The article covers Jev's debut at the AI Engineer Code Summit, introducing creator Diogo Almeida's "automate everything" proposition and the fact that companies like Glean have already deployed Jev in production. Jev's core philosophy centers on System One models—models known for low latency and intuitive judgment, excelling at pattern recognition, fast classification, and simple decisions but unsuited to complex multi-step reasoning. Developers should therefore treat Jev as a component in an automated pipeline for high-frequency, low-complexity tasks such as code formatting, simple bug fixes, log analysis, and routine test generation. The article concludes that judging which tasks suit Jev and which still require human or stronger reasoning models will become a core competency for future engineers.

From Theory to Practice: Jev Is Changing How Developers Work

At the AI Engineer Code Summit, Diogo Almeida, the primary creator of Jev, issued a blunt call to developers: automate everything. This isn't just a slogan—it's grounded in the real capabilities that System One models are now demonstrating. Meanwhile, companies like Glean are already showing how they use Jev—which means the technology has moved from the lab into production environments.

Jev's core philosophy is to use System One models for tasks that require fast responses and intuitive judgment. Unlike System Two models, which require extensive reasoning steps, System One models are closer to human instinct, capable of making decisions at millisecond scale. This characteristic makes Jev excel at automating repetitive engineering tasks, especially in code generation, testing, and deployment.

System One / System Two is a classic pair of concepts from cognitive psychology, popularized by Daniel Kahneman in Thinking, Fast and Slow. System One refers to fast, automatic, intuitive, low-energy thought processes—such as recognizing a face or understanding your native language. System Two is slow, effortful analytical reasoning that requires working memory—such as solving a math problem or doing logical planning. In AI, this distinction has been borrowed to describe two types of model architectures and reasoning paradigms: one pursues low latency and single-step or few-step output, suitable for classification, matching, and pattern completion; the other improves accuracy on complex tasks through chain-of-thought, multi-step planning, and self-verification, at the cost of higher latency and compute consumption. Understanding this contrast helps clarify why Jev emphasizes "excelling at specific tasks" rather than pursuing general-purpose omniscience.

Real-World Application Scenarios for System One Models

Glean's case offers a window into this. Although specific details are limited, it's reasonable to infer that Glean is using Jev to enhance its enterprise search and knowledge management products. The low-latency characteristics of System One models are well suited to search scenarios that require real-time responses, which may explain why Glean chose Jev as part of its tech stack.

Diogo Almeida's talk focused more on how developers can embrace automation. He encouraged engineers to hand off the repetitive parts of their daily work to Jev, freeing up energy to focus on higher-level design and architecture decisions. This approach aligns with current trends in AI coding assistants, but Jev's uniqueness lies in its focus on System One models—not pursuing omniscience, but excelling at specific tasks.

How Developers Can Get Started with Jev

For engineers who want to try Jev, understanding the capability boundaries of System One models is key. These models excel at pattern recognition, fast classification, and simple decisions, but may fall short of System Two models on tasks requiring complex reasoning or multi-step planning. Therefore, the sensible way to use Jev is as a component in an automated pipeline, rather than trying to use it to solve every problem.

Almeida's call to "automate everything" can be understood as a methodology: identify the high-frequency, low-complexity steps in your workflow, then automate them with Jev. This includes but is not limited to code formatting, simple bug fixes, log analysis, and routine test case generation. By handing these tasks to System One models, developers can significantly reduce context switching and manual operation time.

Context switching refers to the cognitive overhead developers incur when shifting attention between different tasks, tools, or mental states. Research shows that frequent switching leads to decreased efficiency, and re-entering a state of deep work often takes minutes or longer. In engineering practice, context switching often comes from interruptive notifications, cross-repository jumps, environment configuration, and repetitive manual verification. Handing high-frequency, low-complexity steps to automated components is essentially about reducing this kind of switching: letting machines bear the cost of "getting into state," while humans retain contiguous time for design and debugging. This is also the direct basis for the "automate everything" methodology at the level of engineering efficiency.

Lessons and Insights from Early Adopters

Glean's involvement shows that Jev has already been validated in enterprise-grade products. Although public information is limited, it's reasonable to speculate that Glean values Jev's advantages in latency and throughput. In enterprise search scenarios, users expect instant results, and the fast response capability of System One models meets this need precisely.

On the other hand, Almeida, as Jev's primary creator, delivered what was more of a call to action than technical evangelism. He didn't overpromise Jev's capabilities, but emphasized that developers should proactively explore automation opportunities. This pragmatic attitude may well be the current hallmark of the Jev community: focused on real output rather than conceptual hype.

Enterprise search, unlike general-purpose search for the public web, targets heterogeneous data sources within an organization—documents, code, tickets, meeting notes, and knowledge bases—emphasizing permission awareness, data connectivity, and result traceability. Users of such systems are typically embedded in a workflow and are especially sensitive to response latency; the difference between sub-second and multi-second responses directly affects adoption rates. Therefore, low-latency retrieval and ranking models are particularly valuable in this scenario, and result recall quality depends on rapid judgment of query intent—which falls squarely within the capability range of System One-style models. Glean's choice of this type of tech stack can be understood as a technical response to a product positioning of "real-time first."

Future Outlook: The Evolution of System One Models

Based on information revealed at the summit, Jev's ecosystem is gradually taking shape. The creator's call and the early adopters' cases together sketch out a direction: System One models will increasingly be embedded in developer toolchains, becoming the default automation layer. For engineers, this means rethinking how they work—not competing with AI, but learning how to effectively direct AI to complete those tedious but necessary tasks.

Of course, System One models are not a panacea. Developers need to judge which tasks are suitable for Jev and which still require human intervention or more complex reasoning models. This judgment itself may become one of the core competencies of future engineers.

#Jev#System One models#Diogo Almeida#AI Engineer Code Summit#Glean#automation#developer tools#enterprise search
Share:

Related articles