Can this AI replace Amazon for smart shoppers?
with Siva, Tonita
Can this AI replace Amazon for smart shoppers?
Show Notes
Siva is the founder of Tonita, an AI-powered shopping search engine at to.co that replaces the tab-overload experience of online shopping with a single conversational interface. Before founding Tonita, Siva spent 15 years at Google as a research scientist working on search algorithms, machine learning, and AI - joining in 2005 after stints in academic research and IBM Research. He left Google in 2021 to build the product he had been frustrated by the absence of as a shopper.
Tonita has raised over $5 million from three venture firms and a group of angels, with South Park Commons as the lead investor. The team is under 10 engineers, all actively building. The product is currently in beta covering clothing, apparel, accessories, and cosmetics - with home goods coming next. It is pre-revenue and optimizing for user growth and engagement quality before monetization. The underlying vision is to become the default AI shopping destination for anyone who has ever opened 20 browser tabs just to buy a $20 item.
What Tonita Does: The Shopping Concierge Model
Tonita is a conversational shopping search engine. You type - or ask - what you are looking for, and the AI guides you through the decision: what factors matter, what the tradeoffs are, what questions you did not know to ask. It shows you real products from real vendors as part of the conversation, not as a separate results page. You can go deeper on any item, follow a link to the retailer, or keep refining with the AI until you know exactly what you want.
Siva's example from his own shopping experience captures the use case precisely: he was looking for a kitchen knife set, ended up with 20–30 browser tabs open across multiple sites, encountered the phrase “Santoku knife,” did not know what it was, opened another tab to research it, realized he needed one, and then had no way to express “I want a knife set that includes a Santoku knife and also a tomato knife” to any existing search interface. Tonita is built to handle that query naturally - the way you would describe it to a knowledgeable friend - and return products that actually match.
The product handles guardrails gracefully too. When asked for Versace knockoffs, Tonita declines to surface counterfeit products but immediately offers Versace-style alternatives at the requested price point - a useful reframe that serves the user's real intent (the aesthetic, not the label) without compromising the platform's integrity.
The Business Model: Transparent Monetization
Tonita's planned monetization has two tracks: paid recommendations and affiliate commissions. The distinction Siva draws carefully is between what is shown because it matches the user's stated preferences and what is shown because a brand paid for placement. His commitment is to make that distinction visible to the user at all times - no mixing organic results with paid links in a way that obscures which is which. He describes this as a line the company is actively drawing, not a future policy.
The paid recommendation layer becomes especially powerful as Tonita accumulates user preference data: if a user has said they prefer sustainable clothing, an alert about a sale on sustainable items in their size is something they genuinely want to see, not an intrusion. That targeting quality is what justifies the premium over standard display advertising, and it only works if the user trusts that Tonita is working on their behalf - which is why the transparency commitment is foundational.
Why Tonita Stayed Broad Rather Than Going Niche
The temptation to niche down was real. Tonita explored anchoring to sustainable clothing, urban women's apparel, and other focused categories before choosing to go broad within clothing and accessories instead. The reasoning behind staying broad is multi-layered. First, too narrow a category fails the “toothbrush test” - it is not something a large enough audience thinks about frequently enough to build a habit around. Second, and more interesting from an AI standpoint: if you niche down too early, you only learn how one narrow demographic uses your product. Tonita needs to understand how people of different ages, income levels, education backgrounds, and geographies shop conversationally - because the answer varies enormously, and those learnings are what make the AI better.
The solution was category restriction rather than audience restriction: cover clothing and accessories for everyone, rather than all of e-commerce for a specific demographic. As the product proves itself, adjacent categories (cosmetics, home goods) are being added. The goal is eventually to be the universal shopping destination - but the path there runs through being excellent at a specific slice first.
From IBM Research to Google to Founding Tonita
Siva grew up in India, did his graduate work at the University at Buffalo (a Buffalo Bills fan for life, present for all four Super Bowl losses), and spent years in academic and industrial research before the internet era pulled him toward search. He joined Google in 2005 as a research scientist focused on search algorithms and ranking - the mathematical questions underlying how to surface the most relevant result - and stayed for 15 years as language understanding and computer vision steadily improved.
The introduction to computing came in 10th grade, at a tiny air-conditioned room in a small town in India where a BASIC programming class was offered over the summer. One hour of computer time per week meant everything had to be worked out on paper first. That constraint, combined with a growing appetite for science fiction (Star Trek was the closest fit; Harry Potter for the lower-sci-fi end of the spectrum - magic grounded in human values), wired Siva toward the idea of systems that seem to think. He left Google in 2021 when it became clear the technology was finally ready to build the shopping experience he had been frustrated by for years.
Siva's Four Founder Mistakes Worth Avoiding
In a short segment recorded separately, Siva offered the most operationally specific advice of the conversation - a direct list of mistakes he made and would not repeat:
1. Raise money only after you have something working. Think through the idea first. Prototype it. You might need an angel or seed round to get initial traction - but do not raise serious capital before the idea has been validated in any form. The pressure of significant capital can entrench decisions before they are ready to be entrenched.
2. Do not rabbit-hole into design before the idea is clear. Siva invested significant energy and money into design detail in Tonita's early days - pixel-level decisions - before the core product concept was settled. That work was largely wasted. Nail the idea first.
3. Hire for ambiguity tolerance, not just skill. The most important hiring question for a startup that often gets skipped: can this person handle uncertainty? Startups have incomplete information, changing priorities, and no playbook. Some excellent engineers cannot operate well in that environment. Ask the question explicitly before extending an offer.
4. Double down on design at the right moment - not too early, not too late. Tonita launched beta with minimal design. That was acceptable for a beta. But for a consumer product, design quality determines whether users trust the product and return. The right time to invest heavily in design is when the core experience is stable enough to be worth polishing - and missing that window is as costly as investing too early.
Tools & Resources Mentioned
- Tonita / to.co - Siva's AI shopping search engine; conversational interface for clothing, accessories, cosmetics, and home goods
- South Park Commons - lead investor in Tonita's seed round; “minus-one-to-zero” program for founders who know they want to start something but are still figuring out what
- Perplexity / Aravind Srinivas - Siva's cited example of a founder with a clean, focused AI search thesis who executed against it with exceptional speed
- The Toothbrush Test - Larry Page's framework for evaluating product utility: does the user think about it and use it at least once a day?
- AngelList - referenced in context of VC power law investing data
- Meta Ad Library - referenced in adjacent episode context
Frameworks
Category Restriction vs. Audience Restriction
When deciding how to focus a broad consumer product, choosing which categories to cover (clothing, not all of e-commerce) is different from choosing which users to target (women 25–34). Category restriction limits scope while keeping the user base diverse - which is valuable when the AI still needs to learn from a wide range of shopping behaviors. Audience restriction limits scope and learning simultaneously, which is often the wrong trade-off early.
The Toothbrush Test for Market Sizing
Larry Page's framework: does a product get used at least once a day by a large number of people? Tonita used this to reject overly niche category ideas (sustainable clothing alone fails the test) in favor of broader categories that create genuine daily-use habits.
Transparent Recommendation Architecture
When a platform shows both organic results and paid placements, users must always be able to tell which is which. Mixing them destroys trust and ultimately destroys the value of the paid placements too - because users learn to discount everything. Tonita draws an explicit line between AI-matched organic results and clearly labeled paid recommendations.
AI as Overnight Success, 30 Years in the Making
Unlike most technology cycles (evolutionary, predictable year-to-year), the AI revolution represents a genuine discontinuity - primarily because it works in both directions simultaneously. Computers now understand humans at a qualitatively new level of fluency, AND they communicate back in a way that is organized, concise, and contextually appropriate rather than just pointing to links. That bidirectionality is what makes this cycle different from the internet, mobile, and cloud.
Hire for Ambiguity Tolerance
Technical skill and startup-environment compatibility are different things. The most important unasked question in startup hiring: can this person operate effectively under uncertainty, changing priorities, and incomplete information? Engineers who excel in structured environments with clear specs often struggle when the spec is ‘figure it out.’ Ask directly, in the interview, with specific scenarios.
FAQ
What is Tonita and how does it work?
Tonita is an AI shopping search engine at to.co. Instead of entering keywords and scanning a results page, you have a conversation: tell the AI what you are looking for, ask questions, get educated about product categories, and refine your preferences - all while seeing real products from real vendors as part of the conversation. You can click through to any product or go directly to the retailer to buy. The AI remembers preferences you express and can surface relevant recommendations later.
Who is Tonita for?
The primary target is adults aged 20–40 who value their time and have active shopping needs across multiple categories. Siva specifically chose not to segment by gender or micro-demographic because the product needs to learn from a broad range of shopping behaviors. Gen Z and younger millennials tend to be the most natural users - they have grown up on text-based interfaces and are comfortable expressing needs conversationally rather than through keyword search.
How does Tonita make money?
The planned model has two components: clearly labeled paid recommendations (brands paying to surface products to users who have explicitly expressed relevant preferences) and affiliate commissions on purchases made through the platform. The core commitment is that organic AI-matched results and paid placements are always visually distinct - users can always tell which is which. This transparency protects user trust and, by extension, the value of the paid placements.
What Tonita will not do?
It will not surface counterfeit or knockoff products. When asked for Versace knockoffs, it declined but immediately offered Versace-style alternatives in the requested price range - serving the user's real intent (the aesthetic) without the brand name or the ethical problem. This behavior is an example of the guardrail work that goes into making the AI behave in a way that is both useful and trustworthy.
Why did Siva leave Google after 15 years to build a shopping engine?
Two things converged. First, as a researcher watching language understanding and computer vision improve over 15 years, he could see that the technology was finally capable of something genuinely new in consumer search. Second, as a shopper, he was routinely opening 20–30 browser tabs for a single purchase, learning new product terminology mid-research, and having no way to express complex requirements (like ‘a knife set with a Santoku knife and a tomato knife’) to any existing interface. The personal frustration and the professional trajectory pointed at the same gap.
What was the South Park Commons fellowship?
South Park Commons is a San Francisco-based community and program for founders at what Siva describes as the minus-one-to-zero stage: they know they want to start something but have not yet locked in what. Siva spent a quarter there while developing the Tonita concept, which expanded his network and eventually led South Park Commons to become Tonita's lead seed investor.
What are Siva's top mistakes for founders to avoid?
Four: (1) Raise serious capital only after you have something working - prototype first. (2) Do not invest in detailed design before the core idea is settled - that work will mostly be thrown away. (3) Hire for ambiguity tolerance, not just technical skill - ask explicitly whether a candidate can operate well under uncertainty and changing priorities. (4) Double down on design at the right moment - too early is wasted, too late costs users. For a consumer product, design quality determines whether people trust and return to the product.