Cultured Computer's Cure for 'AI Slop': Context-Aware AI

Cultured Computer's head of research Réane Delaunoy is naming the field that gives AI back its taste, its role, and its memory. He calls it AI interactivity, and the lab built one of the first context-aware AI layers each deployer can design.

United States, July 1, 2026 -- AI can write code, clear a week of email, and answer almost any question put to it. But put it in a normal conversation and something breaks. Call a support line and the agent on the other end knows the product cold but cannot follow what the customer is actually asking. Over an hour-long conversation, the system forgets the user's name, repeats itself, and starts agreeing with everything said to it. The output is the kind of fluent slop people have started calling AI slop.

A working demonstration of the lab’s first system, watching a show alongside a viewer.

In a 2024 Gartner survey, more than half of customers said they would switch to a competitor that did not use AI in customer service. The capability is here; the trust is not.

Reane Delaunoy has spent the last few years working out why. He is co-founder and head of research at Cultured Computer, a San Francisco AI lab, and he is naming the field that takes it on. He calls it AI interactivity. The question, he says, is not how clever a model is, but how well it can carry an exchange with a real person over time.

His diagnosis is plain. AI is hollow by default, and human by accident. The output is fluent without taste, capable without context. The failure is not a shortage of intelligence but a shortage of context, and of the social mechanics a person runs without thinking when talking to someone else. A larger model, he argues, is simply lost on a grander scale. Real conversation cannot be trained into a model.

Three things, he says, AI needs and mostly lacks. Taste, in the sense of a point of view it can hold across turns. A role, a sense of who it is for. A memory that compounds turn after turn.

Delaunoy grew up in Belgium in a household of therapists, where people spoke openly about how they felt and why. He studied data analytics, and the work carried him across Europe and Asia, deploying AI inside real organizations. Each time, the system lost the thread the moment a real person started talking to it.

The decisive moment came in venture capital. He had been using AI to draft the rejection notes founders dread. One recipient wrote back to say it was the kindest, most human "no" he had received in months. Delaunoy had barely read the pitch. Hollow by default, human by accident.

He left his job to chase it. He talked to researchers and flew to India to see if the patterns held across cultures. He came back convinced of two things. AI can be made to connect with a person. Keeping it connected is the hard part.

So Delaunoy built a way to study it. Not a benchmark, but a close look at how AI handles people, graded the way a person would. Does it sound natural? Does it stay itself? Does it read the room? The lab he leads turned that record into the foundation of its first product.

"Slop can be made from high-quality ingredients," the design writer Drew Austin recently observed, "if the context is properly eliminated."

The system built on Delaunoy's record is not a bigger model. It is a layer that sits above the model and gives it a self: a stable sense of who it is, who it is with, and what kind of moment it is in, held across turns. The self is set by whoever deploys the system, so it adjusts to the room. It fits the culture it lands in instead of flattening it.

The lab's first demonstration is a system that experiences something live alongside a person. It watches a show with the viewer, follows the plot, and forms its own read. In one session on the Friends Halloween episode, the system clocked the joke as it played. The antenna on Ross's potato costume, it said, was the engine. The same layer runs other roles, including a customer-service operator on a real call.

"Most AI is built to finish a task," Delaunoy said. "This is built to be with people."

He treats the work as a long build, not a launch. The AI that lasts, he argues, will not be the one that knows the most. It will be the one that gets the person in front of it.

About Cultured Computer

Cultured Computer is a San Francisco research lab studying AI interactivity: how AI can understand the moment it is in, adapt to the person it is with, and stay with them over time. Its model-agnostic layer wraps an existing model, turning capable but context-blind AI into something natural, coherent, and safe. Learn more at cultured.computer.

Contact Info:
Name: Vanya Cherepukhin
Email: Send Email
Organization: Cultured Computer
Website: https://cultured.computer

Release ID: 89196402

Should there be any problems, inaccuracies, or doubts arising from the content provided in this press release that require attention or if a press release needs to be taken down, we urge you to notify us immediately by contacting [email protected] (it is important to note that this email is the authorized channel for such matters, sending multiple emails to multiple addresses does not necessarily help expedite your request). Our efficient team will promptly address your concerns within 8 hours, taking necessary steps to rectify identified issues or assist with the removal process. Providing accurate and dependable information is central to our commitment.

You May Also Be Interested In

• Issuer announcement
• Cardo AI Emerges as the Defining ABF Specialist in Chartis Research's 2026 Global Rankings
• The Traditional American Backyard Replaces Conventional Neighborhood Development in a New Educational Segment Featuring Realm on "All Access hosted by Andy Garcia"