If models have reached superhuman intelligence, where is all the automation?
While consumer-facing applications of artificial intelligence have captured the public imagination—sparked initially by the viral emergence of ChatGPT and later advanced by specialized developer tools like Claude Code—a fundamental paradox has continued to puzzle industry observers. Despite extraordinary leaps in the raw capabilities of large language models, the everyday software running our digital infrastructure largely operates without native, autonomous intelligence. The massive leaps in reasoning have not yet translated into the seamless, hands-off automation that many industry forecasters anticipated years ago.
Enter TypeSafe, a rising player in the infrastructure space that is stepping forward with a new paradigm designed to solve this exact bottleneck. Formally introducing what it calls the "Jev moment," the company is targeting a shift away from conversational, human-in-the-loop workflows toward models that power software natively at runtime. Instead of waiting for human prompts, these new systems are built to make millions of instantaneous decisions completely behind the scenes, effectively removing humans from the routine loops of software execution.
The market response to this approach has been swift and striking. Within twenty-four hours of its official debut on September 15, TypeSafe’s flagship model, Jev, recorded the fastest adoption rate of any model in the history of Vercel’s AI Gateway. Over the subsequent weeks, early-stage investors and enterprise stakeholders alike reported a massive rush of portfolio companies racing to deploy the technology into production environments for a wide spectrum of newly unlocked use cases. Industry observers note that the sheer volume of developer pull surrounding the launch represents a level of enthusiasm not witnessed in the ecosystem since the groundbreaking arrival of GPT-3.
Decisions Instead of Strings
To understand the engineering breakthrough behind TypeSafe’s approach, one must look at how Jev fundamentally flips the traditional large language model architecture on its head. Conventional language models are designed to generate text sequentially, writing out answers word by word in a slow, deliberate cadence. Borrowing from cognitive psychology, this process is frequently likened to human "System 2" thinking—analytical, careful, but computationally expensive and sluggish.
Jev, by contrast, is engineered to operate as AI’s "System 1": fast, cheap, highly intuitive, and optimized specifically for the thousands of minor judgment calls that standard software applications must execute every single second. Rather than outputting conversational strings of text, the model ingests a rich context and a specific set of operational questions, returning structured, typed answers complete with confidence scores all at once.
This architectural shift unlocks a wide variety of practical applications across the software stack. Developers can deploy Jev to determine precisely which button an automated digital agent should click next, grade evaluation metrics, route incoming network requests, triage technical support tickets, or extract complex data fields from legal contracts. The system delivers an out-of-the-box generality that requires no custom-labeled training datasets or dedicated machine learning engineering teams, making advanced decision-making accessible to standard application developers. Consequently, early adopters have leveraged the technology for an eclectic array of experiments and utilities, ranging from automated gameplay in classic titles like Doom to the dynamic re-ranking of personalized news feeds and the generative construction of multi-dimensional virtual environments.
Jevons Paradox, by Design
The nomenclature of TypeSafe’s core offering is intentionally derived from classical economic theory: Jevons paradox, which observes that as technological advancements make a resource dramatically cheaper and more efficient, the overall consumption of that resource increases exponentially rather than declining. TypeSafe is placing a calculated wager that the economics of artificial intelligence will follow this exact historical trajectory.
When classification, data extraction, and decision-making tasks are economically expensive, engineering teams are forced to ration their usage. They typically resort to evaluating only a small randomized sample of automated agent runs rather than auditing the entire operational pipeline. Alternatively, they rely on brittle, hardcoded software rules instead of querying a model, or they bundle multiple distinct operational questions into a single broad query to save on computational overhead. When the cost of classification drops to near-zero, however, those limiting tradeoffs vanish entirely.
This economic transformation fundamentally alters what is economically viable to build. Many of the most persistent operational challenges across various industries are structurally suited for high-volume, low-cost automated decision-making—a reality underscored by the immediate explosion of creative use cases discovered by early developers. Industry analysts suggest that the most transformative applications built on top of infrastructure like Jev are likely still conceptual, waiting to be imagined as developers grow accustomed to abundant machine intelligence.
Prod, Not God
The foundational vision for TypeSafe originated with Diogo Almeida during his tenure at OpenAI, where he contributed to the development of InstructGPT and reinforcement learning from human feedback (RLHF). During that period, the overarching industry focus was heavily centered on training models to become exceptional conversational partners—optimizing them for direct human interaction and supervised, human-in-the-loop workflows.
However, Almeida recognized that true, scalable enterprise automation required an entirely different set of operational characteristics: reliability, extreme speed, and low cost. His core realization, which he frequently describes as the "bitterest lesson" of modern machine learning development, was that selecting the precise, practical task to solve holds significantly greater long-term importance than merely scaling up model size.
Almeida brings a distinctive background to the artificial intelligence landscape. Born in the Philippines and formerly ranked among the top thirty North American players in the competitive video game League of Legends, he combines the creative, iconoclastic mindset of a foundational researcher with practical developer intuition and sharp commercial instincts. Observers note that this pragmatic yet ambitious profile aligns closely with the leadership qualities required to navigate the transition into the next machine age.
While Jev has successfully captured early market momentum, TypeSafe views the model merely as an opening salvo. The company’s long-term strategic objective is to establish itself as the foundational infrastructure layer governing how modern software makes automated decisions. By creating a robust platform of composable primitives, TypeSafe aims to provide developers with building blocks as fundamental and ubiquitous as traditional databases once were.
In alignment with these expansion goals, TypeSafe has formally closed its Series A funding round, partnering closely with Almeida alongside co-founders Sasha Sheng and Erik Spock Gafni, following an initial investment from a16z.
Comments (0)