The Applied Intelligence Designer

Here is the idea in one sentence: because AI products are profoundly technical, and their behavior is shaped by that technicality, building the best product experience now requires someone who understands the technical layer deeply enough to design for it. A practitioner who bridges product thinking, model operations, and the systems that let AI do reliable work.

The broken contract

Traditional design rested on abstraction. You didn’t need to know how the database was indexed to design a good search box. Deterministic software hid its implementation behind the interface, and the interface was the designer’s canvas.

AI breaks that contract. A model’s technicality is the experience. Why it hallucinates here but not there, why it loses coherence past a certain context length, why the smaller model feels subtly dumber: these are not hidden implementation details. They surface directly as the product’s personality and its failure modes. The abstraction layer that let designers ignore the technology is gone, and the craft now has to span both sides.

The role isn’t speculative; people keep arriving at it from different vantage points. Jason Warner at Poolside argues that intelligence is infrastructure on par with electricity, and that companies which don’t own their intelligence layer end up tenants on someone else’s land. Will Manidis, in “Tool Shaped Objects”, warns that this generation of AI systems is the most sophisticated machinery for producing the feeling of work ever built, and that the discipline is making sure it produces real value instead of apparatus. OpenAI’s “Harness Engineering” write-up, building on Mitchell Hashimoto’s coinage, describes the engineering job splitting in two: building the environment and managing the work. And Jensen Huang’s five-layer cake runs energy, chips, infrastructure, models, applications; this role lives between the last two layers, the connective tissue where intelligence becomes product.

From below, practitioners keep hitting the same wall: teams with capable coding agents stall at the stage of reviewing machine output all day, and the block is a design failure, not a capability failure. The market has even started naming the engineering half of the role; the Forward Deployed Engineer is now one of the fastest-growing jobs in enterprise AI. The design half is still unclaimed.

The third interface paradigm

GUI was built for direct manipulation. CLI was built for precise instruction. Both assume you do the work. The emerging third paradigm is the delegation interface, where you are the manager and the interface is the delegation layer. It has five interaction patterns: commissioning, monitoring, intervention, review, and governance. Designing these well is the defining problem of the role, and almost nobody is working on it seriously.

There is a quieter architectural question underneath: state. When AI becomes operationally embedded, four kinds of state accumulate: behavioral, memory, organizational context, and human-AI working patterns. Whoever owns that state owns the moat. The Applied Intelligence Designer designs systems where the model stays an interchangeable, stateless reasoning engine and the state stays owned and portable.

The moat is the span

Pure ML engineers can’t define what “good” means from the user’s side, or translate model behavior into product decisions. Pure designers can’t fine-tune, serve, or harness a model. The Applied Intelligence Designer does the whole loop: decide what to build, build it, serve it efficiently, design the delegation interface, harness it for reliability, govern the state, and speak to the business about all of it.

Vanishingly few people can operate across that full span today. That gap is the opportunity.