Building AI Systems That Can Be Trusted: Harsh Verma on the Future of Intelligent, Secure Technology
Artificial intelligence is entering a new phase. It is moving beyond systems that simply respond to prompts towards ones that can reason, make decisions, collaborate with other systems, and operate with increasing autonomy. That evolution has the potential to reshape industries ranging from cybersecurity and healthcare to finance and enterprise operations. It also raises a more fundamental question. As AI becomes more capable, how do we ensure it remains dependable?
For Harsh Verma, Principal Software Engineer – AI at Palo Alto Networks, the answer starts by looking beyond the race to build increasingly powerful models. With experience spanning artificial intelligence, cybersecurity, software engineering, distributed systems, and AI agents, Verma views the future through the lens of systems engineering. The challenge, he believes, is not simply creating intelligent software, but designing systems that can operate safely, reliably, and responsibly in an unpredictable world.
From Reactive Systems to Systems of Intelligence
Verma’s interest in AI grew out of years spent building complex, large-scale systems. His early work centred on distributed systems and real-time data platforms, where scale, latency, reliability, and resilience were constant priorities. Those experiences shaped a principle that still guides his work today: technology should not only function under ideal conditions, it must continue performing when real-world complexity intervenes. Over time, however, he began to recognise the limits of conventional software. Even sophisticated systems remained fundamentally reactive. They executed predefined rules efficiently but lacked the ability to adapt meaningfully when circumstances changed.
That limitation became particularly evident in cybersecurity, where threats evolve far faster than static rules can.
“The challenge is not just to make systems powerful, but to make them trustworthy, secure, and aligned with human intent,” Verma explains.
The search for a better approach led him deeper into artificial intelligence and, ultimately, agentic AI—systems designed to reason, coordinate actions, and operate with greater autonomy. To Verma, this represents more than another technological advance. It signals a fundamental shift in the way software itself is designed.
“We are moving from AI as a passive tool to AI as an active participant,” he says.
The distinction is significant. Traditional AI may generate an answer or recommendation when prompted. Agentic AI can plan, make decisions, interact with other systems, and adapt its behaviour as circumstances change. In environments where speed and intelligent decision-making are critical, that difference has far-reaching implications.
The Future Will Belong to Intelligent and Secure Systems
Working across software engineering, cybersecurity, and AI has convinced Verma that these disciplines can no longer be treated independently. Each contributes an essential layer to building intelligent systems that can be trusted. Software engineering provides the foundation through scalability, reliability, resilience, and production readiness. Cybersecurity introduces an adversarial perspective, requiring engineers to think not only about how systems function but also how they might fail, be manipulated, or exploited. Artificial intelligence adds adaptability, but it also introduces uncertainty.
Together, they create a far more demanding engineering challenge.
“You cannot build powerful AI without strong engineering foundations, and you cannot deploy it responsibly without embedding security at its core,” Verma says.
That principle becomes even more important as AI systems gain greater autonomy. The more decisions a system can make independently, the more important it becomes to understand how it will respond to unexpected inputs, malicious attacks, or situations beyond its original design assumptions. The objective is not to eliminate uncertainty altogether. With probabilistic systems, that may never be possible. The challenge is to build systems that can operate safely despite it.
For Verma, that means investing in observability, monitoring, fail-safes, guardrails, validation layers, and human oversight. Intelligent systems must do more than perform well. They should also be measurable, understandable, and controllable.
Trust Is a System-Level Responsibility
One of the biggest misconceptions surrounding AI, Verma believes, is the assumption that it will simply replace human work. The more profound shift is likely to be structural. The future of work, he suggests, will increasingly move from doing to orchestrating. Instead of completing every task themselves, people will supervise intelligent systems, guide their actions, interpret their outputs, and collaborate with them.
That evolution demands a different understanding of trust. AI should not be viewed as perfect intelligence, Verma argues, but as probabilistic assistance. It can deliver extraordinary value, but it also has limits. Knowing when to rely on AI—and when to question its conclusions—will become an increasingly valuable skill across industries.
Trust, however, cannot depend on the model alone.
“Trust in AI will not come from the model alone, but from the system around it,” he explains.
That surrounding system includes transparency, security, observability, governance, feedback loops, and thoughtful human-in-the-loop design. In high-stakes environments such as cybersecurity, healthcare, and finance, human judgement will remain indispensable. The strongest AI systems, Verma believes, will not necessarily be those that operate with complete independence. They will be those that combine machine intelligence with effective human oversight.
Responsible AI Requires More Than Good Intentions
As businesses accelerate AI adoption, Verma believes one of the greatest risks is treating implementation purely as a question of technical capability. The conversation cannot stop at asking what AI can do.
It must also ask how it should behave. Responsibility, in his view, has to be designed into AI systems from the outset. Security cannot be bolted on after deployment, and accountability cannot begin only after something goes wrong. Models evolve, behaviours drift, and new risks emerge continuously.
Responsible AI is therefore not a milestone but an ongoing discipline. Companies need clear ownership, continuous evaluation of system performance, mechanisms for identifying emerging risks, and processes that make failures understandable when they occur. As autonomous systems become more common, particular attention must be paid to threats such as prompt injection, data leakage, model manipulation, and adversarial attacks.
For Verma, the organisations that succeed with AI will not necessarily be those that move first.
“They are the ones that build with intention,” he says.
The same philosophy shapes the way he advises startups and early-stage founders. Novelty alone rarely impresses him. He is more interested in the quality of the thinking behind an idea. Does the founder truly understand the problem? Can the solution scale? Is the team willing to challenge its own assumptions?
His advice is straightforward.
“Don’t confuse capability with value.”
Just because AI makes something possible does not automatically make it useful. The more important question is whether it genuinely improves outcomes for the people using it.
Agentic AI and the Next Generation of Technology
Among the developments that excite Verma most is the rise of agentic AI. He sees autonomous agents as part of a broader shift towards systems of intelligence, where humans, software, and multiple AI agents work together dynamically. In cybersecurity, such systems could detect and respond to threats in real time. Within enterprise operations, they could coordinate complex workflows that currently require constant human intervention. Across many other industries, they could support decision-making while adapting continuously to changing conditions.
Yet those opportunities come with equally significant responsibilities. As AI systems become more autonomous, the need for security, alignment, and trust only becomes greater. The future will not be defined simply by who develops the most powerful agents, but by who can govern them responsibly.
This, Verma believes, is where the most meaningful work lies ahead: building intelligent systems that are not only adaptive but consistently dependable.
Building the Future Requires Learning How to Learn
For those beginning careers in AI and technology, Verma encourages them to look beyond the latest trends.
The fundamentals still matter. Problem-solving, programming, data literacy, and systems design remain essential because the tools themselves will continue to evolve. The ability to learn, adapt, and think critically will outlast any single framework or programming language. He also encourages young professionals to build rather than simply consume. Courses and theory provide a foundation, but genuine understanding comes from designing systems, experimenting, testing ideas, and discovering how technology behaves outside controlled environments.
Perhaps most importantly, he believes people should develop their own perspective. The future of AI will not be shaped only by those who follow every technical breakthrough. It will also belong to those willing to ask more thoughtful questions about which problems deserve solving, what intelligent systems should do, and how they ought to be used.
Designing Intelligence With Responsibility
Harsh Verma’s perspective reflects a broader shift taking place across the technology industry. Artificial intelligence is becoming more capable, more autonomous, and more deeply woven into the systems that shape modern life. The question is no longer whether intelligent systems will transform the future. That transformation is already underway.
The more important question is what kind of systems we choose to build.
For Verma, the answer lies in combining intelligent software with engineering discipline, cybersecurity, human judgement, and long-term accountability. The future belongs not simply to those who create smarter systems, but to those who recognise that intelligence without trust is ultimately incomplete.
As AI moves from responding to acting, the responsibility placed on its creators will only grow. Building systems that can reason is a remarkable achievement. Building systems that can be trusted to reason, adapt, and act responsibly in an unpredictable world is the challenge that will define the next chapter of AI.
