CHAPTERS
- 0:04 – 1:59
Giga’s wedge: winning DoorDash with a product-first support agent
Harj introduces Giga and immediately probes how they stood out in a crowded AI support-agent market. Varun and Esha explain that DoorDash evaluated many vendors, and Giga won by shipping a product that could be deployed rapidly rather than requiring lengthy, forward-deployed consulting.
- •DoorDash evaluated 20+ vendors and chose Giga on merit
- •Key differentiation vs consultative/Palantir-style deployments: product that goes live fast
- •Time-to-live: Giga launched in ~1 week vs months for others
- •Enterprise scale demands broad coverage quickly, not use-case-by-use-case rollouts
- 1:59 – 2:56
“No custom work”: how Giga productizes enterprise complexity
Harj challenges the usual assumption that enterprise support requires heavy per-customer customization. Esha describes Giga’s internal rule: no bespoke builds—anything created for one customer must become a reusable core feature.
- •Strict policy: no custom builds per customer; everything becomes core product
- •Focus on general primitives that can handle both common and edge-case workflows
- •Product approach is positioned as the key to scaling across many enterprises
- •Complex enterprise support is harder than it appears at high volume
- 2:56 – 5:16
The “AI forward-deployed engineer”: Python as a first-class product primitive
Esha explains the bet that makes Giga’s platform general: letting customers express business logic that the system can turn into executable code. By making Python a first-party capability inside the product, Giga replaces much of the traditional forward-deployed engineer role with an AI that generates and adapts logic.
- •Python is embedded directly into the product with controlled “where code can run” primitives
- •Ops teams can describe outcomes in natural language; AI generates the necessary code
- •Giga converts customer artifacts (e.g., large JSON configurations) into instructions + code
- •This approach aims to generalize across customers without bespoke engineering
- 5:16 – 6:34
Founders’ technical edge: model fine-tuning, long context, and cost optimization
The conversation shifts to the founders’ research background and why it matters operationally. They describe deep experience with fine-tuning and context-length extensions, and how that informs what they automate, what they don’t, and how they manage token-heavy workloads cost-effectively.
- •Pre-ChatGPT experience working with transformer models and fine-tuning
- •Claimed milestone: extending Llama 2 context length (4K → 32K) and strong performance
- •Knowing model failure modes informs product boundaries and reliability
- •Token-intensive systems require aggressive per-token cost optimization
- 6:34 – 9:40
DoorDash “multi-party” support: parallel calls, sub-agents, and fraud-aware workflows
Varun walks through a concrete DoorDash workflow that other vendors couldn’t solve: verifying delivery when the driver is outside a geofence. Giga’s agent coordinates parallel calls—talking to the Dasher while simultaneously calling the customer—then reconciles context to take action in DoorDash’s systems.
- •Example: address change delivery causes fraud/geofence issues and blocked “delivered” status
- •Agent checks chat history, then runs two calls in parallel (Dasher + customer)
- •Customer verification feeds back into the Dasher call; agent marks the order delivered
- •Protocol evolution: outbound-call “sub-agent” capability inside one agent/session
- 9:40 – 10:36
Why AI can be better than humans: no holds, faster resolution, multilingual accuracy
Harj asks how AI agents can outperform human support beyond cost savings. Varun and Esha emphasize elimination of hold times, faster resolution, handling multilingual and accented speech, and even improved experiences in multi-party scenarios where humans must put people on hold.
- •Eliminates hold times and reduces overall resolution time
- •Handles multilingual conversations and diverse accents at scale
- •Parallel conversations improve experience vs humans who must place callers on hold
- •Reported CSAT improvements (notably for Dashers)
- 10:36 – 13:00
Raising to meet demand: why the Series A now and how pipeline compounds
Harj transitions to funding and go-to-market momentum. Varun explains the Series A rationale—serving a growing queue of large enterprises—and how big contracts plus C-level buyers create referral-driven pipeline expansion.
- •Funding motivation: meet demand from pilots and large pipeline; scale hiring and delivery
- •Enterprise deal sizes pull in C-level decision makers
- •Word-of-mouth among executives drives warm introductions across companies
- •Giga becomes part of the “AI strategy” narrative at the board/executive level
- 13:00 – 14:59
Why customer support is the AI ‘sweet spot’: context constraints and narrow problem scope
Harj asks why support is so competitive. Esha argues that current LLMs excel when the context needed is bounded and can be provided within the interaction—making support (and coding) especially tractable relative to broader managerial work.
- •Support and coding fit current model limitations because relevant context can be scoped and supplied
- •Enterprises can provide policies, customer history, and case context during the interaction
- •As context windows and tooling improve, more enterprise roles become automatable
- •Competitive intensity reflects immediate ROI and feasibility today
- 14:59 – 18:33
Building the enterprise team: hiring for high agency, IQ, and mission-driven impact
As Giga grows beyond ~20 people, Harj probes culture and recruiting. Varun and Esha describe a bar for exceptional talent, prioritizing high agency and impact motivation, and note that the company can attract candidates even against top AI labs by offering outsized ownership and real-world reach.
- •Hiring filter: extremely smart, high-agency, mission-aligned people
- •Pitch: direct impact on experiences for hundreds of millions (or billions) of end users
- •Ability to recruit against OpenAI/Anthropic via impact and ownership, not pay
- •Company mission expands from support to broader operational execution optimization
- 18:33 – 23:56
Founding story: IIT Kharagpur, YC interview shock, and taking the startup leap
Harj asks for the origin story: how the founders met, their first YC application, and why they committed to startups over lucrative alternatives. They recount applying with an education idea, getting challenged in the interview, and taking the risk despite cultural and family pressure in India.
- •Met at IIT Kharagpur (2019); initially applied to YC with an education/jobs platform
- •YC interview pushed them to consider entirely different ideas
- •Turned down high-paying HFT offers and other prestigious paths; high risk in India’s context
- •Motivations: passion, logical expected value, and a desire to tackle the hardest problems
- 23:56 – 28:54
YC journey and visa setback: remote batch, fundraising dynamics, and finding the right wedge
They describe being unable to enter the US for the batch due to visa rejections, running YC remotely, and how being in Silicon Valley later accelerated iteration. The segment also covers early fundraising realities and the role of network effects and YC support in closing their seed round.
- •Visa interview rejections forced them to do YC remotely and slowed early momentum
- •Eventually arriving in the US helped accelerate idea discovery and execution
- •Seed/Demo Day fundraising felt harder from India; relied on a smaller set of investors
- •YC network and key advocates helped close early capital needed to pivot effectively
- 28:54 – 32:53
From fine-tuning-as-a-service to Giga: pivot logic and the first support deployment (Zepto)
Harj traces their pivot path: they raised on a fine-tuning/inference business, then realized it was structurally disadvantaged as frontier models improved and got cheaper. The turning point came through customer conversations (including Zepto), on-the-ground call-center observation, and a decision to stop doing bespoke consulting and instead build a scalable product.
- •Fine-tuning business had traction but was a “broken funnel” as OpenAI/Anthropic improved price/performance
- •Realization: they were “betting against the AI wave” and needed a business that benefits from model progress
- •Zepto became an early customer; founders observed real call-center operations to learn workflows
- •Early solution was highly custom (even web automation due to missing APIs), prompting a productization push
- 32:53 – 35:56
Long-term vision: a context-rich enterprise automation platform and market consolidation
To close, Harj asks where enterprise agents go as models improve. Esha and Varun argue that data + context will be the durable advantage, enabling consolidation into platforms that expand from support into broader operational domains, with Giga aspiring to be the foundation for future large businesses built on top.
- •Strategic belief: context and enterprise data are the key moat, not just model intelligence
- •Support is the beachhead that yields rich customer and operational context for other workflows
- •Expectation of consolidation into broader platforms vs many narrow point solutions
- •Inspiration from companies like Salesforce (generality) and leading AI labs (innovation and taste)
