Aniket Deosthali

Right now, I’m focused on scaling Envive AI and helping define what the next generation of online shopping looks like.

We’ve already built the foundation, AI agents that can guide shoppers more like a real sales associate, and now the work is about expanding that across more brands, more use cases, and more complex customer journeys. The goal is to make the experience feel more natural, more helpful, and ultimately more effective at driving real business outcomes.

A big part of my time is spent working closely with our partners to understand where traditional ecommerce still falls short, and how our agents can show up across more of the journey, not just on-site, but anywhere shopper intent is happening. That includes earlier touchpoints in the funnel, like traffic coming in from ads, where there’s often a disconnect between intent and experience. We’re starting to extend our agents into those surfaces so brands can better capture that intent and create a more seamless path from discovery to conversion, while staying true to their voice.

At a broader level, I’m thinking a lot about how AI becomes an active participant in the shopping experience, not just something you interact with when you have a question, but something that helps guide decisions from discovery all the way through purchase.

Currently, I’m closest to AI systems that operate directly in the shopping journey, things like sales agents that guide product discovery, search agents that understand natural language, and customer experience agents that handle support in real time.

At Envive AI, these systems are all connected, pulling signals from across the funnel and continuously learning from real customer behavior to improve outcomes like conversion, retention, and discoverability.

I spend a lot of time on workflows where these agents aren’t siloed, but working together and improving through reinforcement learning, using live feedback loops to understand what’s actually driving the right outcomes and adjusting in real time. Over time, that allows the system to become more coordinated, more context-aware, and more effective, rather than just a collection of features.

The use cases I’m most focused on are the ones where AI is directly tied to outcomes, not just information retrieval. Instead of simply answering questions or surfacing products, the system is optimizing for things like helping a shopper make a decision faster, increasing conversion, or reducing drop-off, and continuously learning what works to drive those results.

We’re solving for the fact that online shopping is still too fragmented and effort-heavy, not just for consumers, but for the brands trying to understand and convert them.

For shoppers, finding the right product often means navigating endless options with very little guidance. For merchants, understanding what customers actually want requires stitching together insights across multiple tools and systems, which makes it hard to act on intent in real time.

At Envive AI, we’re building AI agents that close that gap by not only guiding customers through the journey, helping them discover, evaluate, and decide faster, but also capturing that intent and learning from what customers actually choose to buy. That feedback loop is what allows the system to continuously improve and align more closely with real behavior.

The constraint is that all of this has to happen in a way that is accurate, brand-safe, and fully controllable, while operating in real time and driving measurable business outcomes, not just engagement.

What I’ve learned is that getting AI to work in the real world has a lot less to do with the model itself, and a lot more to do with how well it fits into the existing system around it.

The biggest unlock has been treating AI as part of the operating layer, not a standalone feature. That means it has to plug into real data, real workflows, and real business logic, everything from product catalogs and inventory to merchandising rules and brand voice.

I’ve also learned that control and trust matter more than raw capability. Teams need to understand what the AI is doing, be able to shape its behavior, and rely on it to be consistent, especially in customer-facing environments.

And maybe most importantly, success comes down to iteration. The systems that actually work are the ones that are constantly learning from live interactions and improving over time, not the ones that are designed once and left alone.

The clearest operational impact I’ve seen is when AI is embedded directly into decision moments, not just supporting them. When it’s actually helping a shopper choose between products, narrowing options, or removing hesitation in real time, you start to see meaningful shifts in conversion, speed to purchase, and even return rates.

At Envive AI, that impact comes from systems that are continuously learning from live behavior, not just static data, and adjusting how they guide each customer accordingly. That’s where AI starts to feel less like a feature or a gimmick and more like a core part of how the business operates.

Where reality has fallen short is in the expectation that AI alone is the solution. In practice, the hard part isn’t generating responses, it’s making those responses reliable, context-aware, and actually effective at helping convert the customer. A lot of early implementations look impressive but break down when they have to operate at scale, in real environments, with real customers.

One shift I expect over the next 12 to 24 months is that AI will move from being something that assists workflows to something that actually owns outcomes.

Today, most production AI is still scoped to tasks, generating content, answering questions, or supporting decisions. But we’re already starting to see the transition toward systems that are measured and optimized against real business metrics like conversion, revenue, and retention.

At Envive AI, that shows up in how AI agents are evolving from reactive interfaces into systems that continuously learn from live interactions and adjust in real time to improve performance.

As that becomes more common, I think we’ll see AI embedded much more deeply into core operations, with clearer accountability, tighter constraints, and a much stronger link between what the system does and the results it drives.