Intent Driven Design

Weird; I mentioned the Jeevons paradox a couple of weeks back and now it’s all over the geek news. It’s not me, it was the launch of Jev this week. To amateurishly summarise, Jev is attempting to be a new type of logic primitive that has embedded AI in its core. It looks like a classifier of three sub-primitives: fuzzy binary (called a noul), choice and score, but it uses AI to determine the value. Its value proposition is that it is reliable, fast, simple and cheap, which means it can truly be used as a logic primitive within an agentic app.

What caught my eye is not so much what it is, but how to use it. Instead of giving a prompt with as much or little context as you like, you must feed it a decision to be made. Its context is bounded by the sub-primitive that you’re choosing, so if you want it to make a choice you need to give it the list of allowed choices. But you don’t have to be precise about the reasoning to lead to a choice, just about the choices themselves.

The benefits for Jev is that it makes the LLM a little more deterministic in its output and so reliable. This same determinism means less tokens burnt on spurious creativity so it appears to be an order of magnitude cheaper. The disadvantage is that the problem needs to be restated so that it is solvable by applying one of the three sub-primitives. This means no longer vibing out “make me an app that creates a shopping list for next week’s meals” but more “is this item in the list of ingredients for my meals next week? -> add to the shopping list”.

To my eye, this is the same as being clear about the intent (the outcome) without constraining the solution. And because the intent is expressed in terms of one of three primitives, it is easy to test their assertion. So as long as we’re confident in how the intent is coded into the logic, we can test the output of the agentic app, and the chain of intent is proven. Viewed through this lens, it seems a big step towards bringing Intent Driven Design into agentic app development.

As I’ve written before, software creation is already pretty much on-demand; every new whim can be turned into deployed code within days or hours. Because code is cheap, the temptation is to not think hard about what problem you’re solving, fulfill every whim, create slop. The real challenge of this agentic moment is to not add to the slop; instead to be very selective about the problem to solve up front, then persist this intent through the ADLC. We will need tools like Jev to encode and validate the chain of intent as code. Such a toolset will enable us to follow Intent Driven Design.

Want more? Download The New Agentic Organisation at https://www.sinharay.tech/free-book.

© Mind Rocket Services Ltd 2026. All rights reserved.

Avi Sinharay

CEng, MIET, MEng, MA (Cantab.)

Fractional CTO, Director/VP of Technology

Core Expertise

Technology Leadership, AI Native Dev, Operating Model Design, Engineering Culture

Domains

Health Tech, Media Tech