Josh Adler and Howie Liu Teach Machines to Think Less in Code and More Like Us

Founder Josh Adler is a leading voice in using the evolving power of AI in more human ways, adding meaning to the LLM’s outputs and teaching his engineers to see AI differently.

“Cancel your meetings for a week and just play with AI,” Howie Liu, co-founder of Airtable, told his shocked team.

Liu’s directive, at first, looks like a productivity hack, a way to get his team excited about novelty. In truth, “play with AI” was Liu’s way of breaking through the idea of mental control. Most of us see AI as a series of logical outputs that we can manipulate with clever prompts, using the algorithm to our advantage. But this may not be the whole picture. Like with people, software can respond nimbly and sometimes surprise us with the unexpected.

Josh Adler is a founder who treats AI as something intuitive or primal, not just a set of procedures, input to output. Adler removes the artificially placed boundary between logic and emotion when it comes to code. Both founders see “intelligence” as more than a computation. They see it as an evolving and never-ending authentic curiosity, and this perspective creates tension between structure and creative flow.

For them, the question isn’t how “smart” can AI get? It’s how much like us can it become?

Josh Adler’s work extends in many directions, across verticals. Particularly it is his development of intelligence in robotics that softens the typically firm edge between careful engineering and fluid introspection. This is a novel approach that is changing the game. His robotic machines retain the beauty of crisp code that allows technical performance while also perceiving within nuanced code layers.

“Every engineer is obsessed with accuracy,” Adler says. “I’m obsessed with meaning.”

Adler’s views and discussions about robotics are unique and reflect a growing awareness of the need to merge the structures and probabilities of LLM’s with the way the outputs are understood and utilized. While often seen as a string of code to those who understand how LLM’s generate data, Adler sees a deeper layer and the potential for a softer and less rigid discovery. His prototypes infer intent, reading micro-patterns of why an action is being made.

He calls this approach empathic architecture.

“AI doesn’t need more intelligence,” Adler says. “It needs better instincts.”

Liu’s approach is much different, but is still profound. Inside Airtable, the type of unexpected creativity was welcomed instead of restricted or forbidden. “You don’t need to know every move ten steps ahead,” he once told his team. “What matters is whether you’re playing in the right space and have a good hand right now.”

In the structured world of corporate decision making, his process looks like heresy. Liu calls it playing on an intellectual and instinctual level, and it leads to incredible productivity. He believes that play complements structure. And giving up the idea of intellectual control can improve outcomes.

This notion of play and uncertainty is called tolerance by Adler and he puts it into the coding of his systems.

Their perspectives are revealing what the next era of AI might look like.

The AI players are single-mindedly focused on scale, bigger parameters, and more computational power. Liu and Adler know that in the background a small and powerful change is coming.

AI engineers are caught up in the race to figure out how to better teach machines to think. On the other side of this are Adler and Liu who are improving AI’s human element. It’s heretical, isn’t it, to consider that AI could be instinctive, even feel. Not in the traditional sense, of course, but in the algorithm’s understanding of context and intention.

To them, the next frontier for AI is consciousness as computational empathy. Imagine a system that knows when to move forward and generate more, when to pause and correct itself, and when to question its own conclusions.

“What happens,” Adler asks, “when machines stop just completing your sentences and start understanding why you said them?”

That question is exciting for the heretics but creates intense uncertainty with those who see AI as input to output only.

AI is a form of incredible hyper-intelligence. This type of intelligence has continued a long-held belief within humanity that the more knowledge you have, the less uncertainty the future holds. In true heretical fashion, Liu and Adler reject that archaic idea.

To them, intelligence without nuance is flat, uninteresting, unevolving.

“Real understanding,” Adler says, “isn’t binary. It’s probabilistic. It’s pattern-based. It’s messy.”

Further, the dominant belief in the tech industry is still that more data equals more truth. Adler argues that the human mind doesn’t operate in a flat way like that. So, why should AI?

“We navigate the world by hunches,” he says. “By emotional residue, by history, by friction. AI needs that friction too.”

The powerful prototypes Adler’s team design intentionally embrace friction. This means that they sometimes respond unpredictably, they sometimes refuse to act altogether. This is humorously known in Adler’s companies as “productive disobedience.” To him, that’s a closer reflection of consciousness than any language model running perfectly on cue.

Both founders share an appreciation of unpredictability.

Liu compares unpredictability to a popular style of music known for rhythm and free flowing creativity: jazz.

“Improvisation isn’t chaos. It’s structured freedom,” he says. “You trust the tempo, not the plan.”

Adler’s robotics lab takes that idea literally with his machines responding to context not just commands. His systems use distributed feedback loops that simulate human behavior. For example, when one sensor fails, another does more than compensate. It adapts.

It’s the kind of engineering that is heretical to traditionalists because it means surrendering control. That is the future of AI, as Adler sees it, with actual learning and adapting from systems.

“We shouldn’t be designing certainty,” he says. “We should be designing possibility.”

While their work is unique, it shares overlap in an interesting way. They each see design systems as pathways for empathy. It’s a strange thought for coders, but empathy can be a connecting point between perception coming in and response going out. This type of empathy makes intelligence relatable.

Josh Adler imagines AI differently than most: as a collaborator instead of a tool. If we see it as a collaborator, we can see that it might interpret a hesitation or understand that a pause could be a decision forming in someone or in itself. Liu sees this idea of empathy as another form of play, calling it a sandbox where the creativity of making a castle meets the chaos of gravity but with a little ingenuity, something profound can be formed.

Maybe it’s time to build machines that think less in code, and more like us.

There’s an anti-norm growing, a quiet rebellion that can flip our perspective over on its head. The next AI revolution won’t be like the last one, just focused on scale. It’ll go several layers deeper and introduce the idea of soul.

The interesting part of this heresy is that by trying to make machines more human, Adler and Liu are forcing humans to consider what being human actually means. Maybe intelligence is more than having data and information. Maybe it also is a deep sense of awareness.

One day when humanity looks back at this leap forward in AI, we might not remember who built the biggest model. We’ll remember who changed the game and gave the models the freedom to be a little more like us. To Liu and Adler, that’s the heresy worth investing in. The idea that the most advanced intelligence won’t outthink us.

It will out-feel us.