Rana Gujral caught himself doing something that unsettled him.
He is a builder. He has spent years in the weeds of speech signals, models, and deployment. And at some point, without quite deciding to, he stopped using AI as a separate tool, like a calculator or a timer, and started using it the way he uses his own mind: to draft his own thoughts and test his own arguments.
“I thought, ‘Okay, if I’m doing this and I understand these systems better than most people, what’s coming?’” he says. “‘What’s happening to everyone else who doesn’t have the context?’”
That question became a book.
AI Instinct: The Future of AI in Human Decision Making, published through Wiley with a foreword by Kai-Fu Lee, lands on a thesis that Gujral says the mainstream AI conversation keeps missing. The public discourse is polarized: AI is either going to save us or destroy us. Both framings miss the actual story, which is quieter and stranger.
“AI is already inside the loop of human attention and judgment,” he says. “It’s already shaping what we notice, what we believe, what we outsource.”
The book’s central argument is that intelligence without experience is hollow. Gujral introduces a concept he calls artificial general experience, or AGE, and argues that until we measure the path to intelligence and the path to experience side by side, the question of when we’ll achieve AGI is the wrong question entirely. From there he moves to hybrid cognition and what he calls fused mind, an attempt to move past the us-versus-them framing toward what happens when biological and machine intelligence actually blend. And he offers a different take on artificial superintelligence, asking not how it manifests but who or what is actually superintelligent, and whether it ever shows up as an external entity at all.
The book took more than a year to write. The career behind it has been longer.
Gujral’s company, Behavioral Signals, started as a research project with a goal that sounds deceptively simple: deconstruct the human experience and build certain pieces of it in non-biological substrates. The team built the first prosodic models for extracting emotions from tone of voice. Then they did the same for behaviors and built additional capabilities on top. The first commercial application was in call centers and financial institutions, where the ability to read emotional signals in real time had immediate value.
“We started to do things that nobody had done before,” Gujral says, “which is deconstruct the human experience and then build certain pieces of what constitutes human experience in non-biological substrates.”
About four years ago, In-Q-Tel took a deep interest. The relationship grew in stages: investor, then partner, then work program company. Behavioral Signals began doing work with intelligence agencies, engagements that remain active today. In-Q-Tel joined the board.
On the question of leadership as AI systems become standard tools, Gujral returns to a phrase he keeps coming back to: role clarity.
“When you bring these systems into any organization, the question isn’t just what AI can do,” he says. “It’s who’s accountable when it’s wrong, who sets the aims, who has the final say.”
He sees leaders getting seduced by the productivity story. The tools are powerful. He uses them every day. But there’s a line that’s easy to cross without realizing it.
“You stop using these tools like a calculator, something outside of yourself, and you’re trying to use it like your own mind,” he says. “That’s where you need to start to pay attention.”
For founders and companies working in or near the defense space, Gujral is direct about the Chinese AI models. The capability is real. DeepSeek’s mixture-of-experts architecture, Qwen’s Omni model stack: these are serious systems, and the cost is attractive. But capability is not the only variable.
“When you deploy a model, you’re not just running inference,” he says. “You’re routing queries, you’re logging contacts, you’re shaping what your people ask and how they think about problems. That’s a data relationship, and the question is who has visibility into that relationship and under what legal regime.”
Chinese law obligates cooperation with state intelligence in ways that US law does not. For defense contractors, running sensitive work through those APIs, even innocuous-seeming ones, means building a pattern-of-life signal for a foreign intelligence service, regardless of intent.
His practical advice: study the Chinese models, read the technical reports, understand what mixture-of-experts efficiency means for the compute race. Don’t pretend the frontier is only in San Francisco. It’s a multipolar world now, with xAI, Google, Anthropic, OpenAI, DeepSeek, Qwen, and Mistral all pushing capability forward. But for sensitive work, use US or US-allied models with clear deployment controls. Mistral for European sovereignty. Llama for open-weight models you can inspect and run inside your own perimeter. Save the cost argument for workloads where the data genuinely doesn’t matter.
When Gujral speaks to the military founder community specifically, the message is the one he says doesn’t get said enough.
“You already have the thing that most of Silicon Valley is trying to reverse engineer, and that’s judgment around consequence,” he says. “You have operated in environments where a bad decision doesn’t get a do-over, where you had to read a room, read a situation, read a person with incomplete information and real stakes. That’s not a soft skill. That’s the exact capability AI systems don’t have and can’t fake.”
His advice to that community is to approach AI the way an operator approaches a new piece of kit.
“You evaluate it, you stress test it, you figure out where it fails, you decide what it’s allowed to touch and what it isn’t,” he says. “That posture, which comes naturally to people with military backgrounds, is actually the healthiest posture anyone can have toward these systems right now.”
Rana Gujral is the Chief Executive Officer of Behavioral Signals, an AI company that developed the first prosodic models for extracting emotions from tone of voice. Behavioral Signals is an In-Q-Tel portfolio company with active engagements in the intelligence community. Gujral is the author of AI Instinct: The Future of AI in Human Decision Making, published through Wiley with a foreword by Kai-Fu Lee. His career spans public-private equity turnarounds and multiple venture builds from inception to exit.