The technology landscape has undergone a fundamental shift over the past decade. What once required identifying a clear problem and executing within tight technical and resource constraints now exists in a more layered environment - shaped by artificial intelligence, abstracted systems, and new expectations of speed and scale. The barriers to building have fallen dramatically. In many ways, anyone can now become a builder.
Ishan Malhotra, a specialist in advanced AI based technologies, shares his perspective on this shift is informed by experience across multiple technological eras. Before the recent surge in AI capabilities, he began building at a time when execution itself was the primary constraint. As a teenager, he founded his first company, Pluto, a low-cost IoT and GSM-based device that allowed farmers to control water pumps remotely. He spent more than five years building and scaling distribution across major states in North India which was an early exercise in navigating the real-world limitations of hardware, connectivity, and last-mile reach.
That context contrasts sharply with today’s environment, where execution has become significantly more accessible. The constraint, as Ishan sees it, has shifted. It is no longer about whether something can be built, but whether it should berequiring sharper judgment around problem selection, user need, and long-term relevance.
After studying computer science at the University of Chicago, where he was involved in research and the startup ecosystem, Ishan Malhotra joined Palantir. There, he worked on applied AI systems across high-stakes domains and industries. He later joined an early-stage insurance technology company as a founding engineer, focusing on AI-native systems in a traditionally slow-moving industry.
Having worked across both large-scale and early-stage environments, and across domains ranging from IoT to agentic AI systems, Ishan brings a valuable perspective on the capabilities and limitations of modern AI, and where the next generation of builders can focus.
Q. You've worked across early-stage and large-scale AI environments. How do you see the role of AI evolving?
The role of the engineer is being rewritten.
For a long time, software was a coordination problem. You needed multiple specialists because execution was expensive. That constraint is disappearing. A single person with clarity can now build what previously required entire teams.
But the more important shift is what AI still cannot do.
There is a useful distinction between intelligence and judgment. Intelligence is anything with a clear notion of correctness - coding, data extraction, summarization, synthesis. AI is already strong at that.
Judgment is different. It is taste, intuition, and the ability to decide under ambiguity. What to build, when to ship, what “good enough” means. In real systems, that is still the hardest part.
In practice, many failures in AI systems do not come from model errors. They come from poorly defined expectations across multi-step workflows. The bottleneck has moved from execution to articulation, how clearly a problem is defined so that systems can reliably operate on it.
Q. How does that change what a young builder should be thinking about?
It changes what you optimize for.
There are two types of products emerging. One improves productivity, helping people do the same work faster. The other replaces the work entirely.
Productivity tools are fragile. They can be replicated or absorbed as models improve. Replacement systems are more durable because they compound with better models instead of being displaced by them.
That distinction matters more than it initially appears.
Another shift is simpler: not knowing how to build is no longer a real barrier. It is now possible for someone with a clear idea to ship a working product in a weekend using modern tools. The gap between developer and non-developer is collapsing.
So the question was never really “can I build this?”
It has always been: do I understand the problem well enough to know what to build?
That remains the hardest part.
Q. Where do you see AI’s most practical impact today?
The most immediate impact is the recovery of thinking time.
A large portion of knowledge work is still repetitive; summarizing, formatting, organizing, retrieving information. AI is increasingly able to handle these tasks, which shifts human effort upward.
What remains is judgment: forming opinions, identifying the right problems, deciding what matters.
The bigger shift, though, is structural. AI is moving from a system you consult to a system that acts. That changes how products are built. Interfaces are increasingly designed for agents executing intent rather than humans clicking through workflows.
That introduces new constraints that are often underestimated - latency in multi-step flows, compounding costs in reasoning-heavy systems, and failures that emerge only at scale.
On a personal level, AI is useful for compressing information quickly. It allows faster synthesis across large sets of inputs. But it also removes friction. And without friction, it becomes easier to miss nuance. The best thinking still comes from sitting with ideas long enough to challenge them.
Q. You built Pluto as a teenager. How do you look at that work today?
The problem was simple: farmers needed remote control over irrigation systems.
But the complexity was in execution - hardware reliability, connectivity gaps, and distribution across regions where infrastructure was inconsistent. Building something that works in the field is very different from building something that works in a controlled environment.
That experience still shapes how I think about products.
A lot of the most interesting systems today are about access. AI is making it possible to deliver services that were previously too expensive or operationally heavy to scale.
Increasingly, the opportunity is not software that supports work, it is systems that do the work.
Q. What matters next, both for the industry and for young builders?
Two things are becoming important at the same time.
First, technical literacy.
Founders do not need to be researchers, but they do need intuition for how these systems behave. How inference actually works. Why latency increases in multi-step agent flows. What context windows mean in practice. How evaluation differs in production systems. Where RAG helps and where it introduces failure modes.
Most AI products fail not because models are weak, but because system design assumptions break under real-world conditions.
Without that understanding, it is easy to build systems that look good in demos but fail in production.
Second, the opposite is also becoming more important.
As AI absorbs more execution, what remains is increasingly human. Relationships, trust, and proximity to real problems matter more than before.
Some of the strongest opportunities come not from isolated insight, but from being close enough to people or environments where problems are forming early.
So, the direction is not one-sided. Builders need both: deep understanding of the technical stack and strong grounding in human context.
Both matter more than they used to.
