CHAPTERS
Giga’s wedge: product-first enterprise support agents (and why DoorDash picked them)
Harj introduces Varun and Esha and asks how Giga stands out in a crowded AI support-agent market. The founders explain that their biggest differentiator is a productized approach that can go live fast, rather than a slow, forward-deployed/consulting model.
- •Competed against 20+ vendors to win DoorDash
- •Differentiation vs consultative “Palantir-style” deployments
- •Product focus enables much faster time-to-value (week vs months)
- •Support for complex, real-world enterprise edge cases is a key advantage
Why traditional customer support (and many AI rollouts) fail at scale
They unpack why customer support is deceptively hard at large enterprises: high volume plus a meaningful tail of complex, multi-party issues. Many vendors can handle simple flows, but break down on the hardest cases that matter for trust and broad adoption.
- •Customer support has a long tail of complex cases at DoorDash scale
- •Complexity often involves multiple stakeholders and real-time coordination
- •Many competitors top out at “easy” tickets and struggle to generalize
- •Winning requires solving the hardest cases, not just the majority
“No custom work” as a company rule: building a generalizable core product
Esha explains Giga’s internal rule: nothing is built as one-off custom code for a single customer. Instead, every feature added for a major customer becomes a core primitive usable by all customers, forcing the platform to generalize.
- •Hard rule: no customer-specific bespoke builds
- •Features built for DoorDash must become reusable platform capabilities
- •Complex customer requests drive product primitives, not services work
- •This approach is difficult but compounds across deployments
AI forward-deployed engineer: turning business intent into executable logic
Giga reframes the forward-deployed engineer model: the real job is translating customer business logic into code and configuration. Their bet is that AI can do much of that translation inside the product, enabling non-engineers to implement sophisticated workflows.
- •Core insight: FD engineers translate ops/business logic into code
- •Giga uses AI to convert natural language requirements into implementation
- •Ops teams can drive changes without heavy engineering involvement
- •Positions AI as a built-in “forward-deployed engineer”
Python as a first-class primitive inside the platform (and why that enables breadth)
They describe a key architectural choice: Python is embedded as a first-party capability with controlled primitives for where code can be injected. This makes the system inherently general and extensible across many enterprise workflows.
- •Python is a first-party element within the Giga product
- •Clear primitives define safe/allowed places for custom logic
- •AI-generated Python enables rapid creation of new protocols/use cases
- •Generalization comes from coding primitives, not one-off integrations
Deep LLM expertise as an unfair advantage: context length, cost, and reliability
Varun and Esha discuss their background in fine-tuning and model optimization, including context-length work on Llama. This informs what they choose to automate, how they manage token-intensive workflows, and how they optimize cost/performance for enterprise deployment.
- •Experience with fine-tuning and pushing model context length (e.g., 32K)
- •Pragmatic approach: automate only what models reliably handle
- •Token-intensive workflows require strong cost/per-token optimization
- •Model know-how translates into enterprise-grade performance and margins
DoorDash’s “address change” case study: multi-party parallel calling in production
They walk through a live DoorDash use case involving fraud checks and geofencing that requires coordinating with both Dasher and customer. Giga’s agent can run parallel calls, verify intent, and then take an operational action (marking delivery) in real time.
- •Problem: Dasher can’t mark delivered if outside geofence; fraud systems trigger
- •Agent checks customer–Dasher chat history for address-change request
- •If needed, agent calls customer in parallel while staying with Dasher
- •On verification, agent marks the order delivered—end-to-end resolution
Why AI can outperform humans on support calls: speed, parallelism, language coverage
Beyond cost savings, they argue AI can create a better support experience than humans in certain scenarios. Removing hold times, reducing resolution times, and handling multilingual/accents improve CSAT—especially when parallel conversations are required.
- •Eliminates hold times; faster time to resolution
- •Parallel conversations (no “please hold” while checking with someone else)
- •Multilingual support and accent robustness at scale
- •Reported CSAT improvements for DoorDash Dashers
Raising $61M: scaling to meet Fortune 500 demand and expand deployments
Harj shifts to fundraising: why raise now and what the capital is for. Varun explains they have heavy inbound demand and large pilots, so the round supports hiring and delivery capacity to maintain quality at scale.
- •Capital raised to meet demand and accelerate hiring
- •Many Fortune 500 pilots and a large pipeline require execution capacity
- •Focus on delivering high-quality experiences across enterprises
- •Scaling the team and infrastructure to support rapid rollouts
How Giga built an unusually strong enterprise pipeline: C-level deals and referrals
They attribute pipeline strength to large contract sizes and selling to senior executives. Once results are clear, referrals propagate among C-level networks, and Giga becomes part of the company’s board-level “AI strategy” narrative.
- •Large enterprise contracts pull engagement to C-level decision makers
- •Word-of-mouth spreads through exec networks across large companies
- •Results at DoorDash help establish credibility quickly
- •Customer support framed as a core AI strategy initiative
Building an enterprise-grade team: high agency, raw IQ, and outsized impact
They discuss hiring and culture as they scale from ~20 people. The founders emphasize selectivity, mission alignment, and the appeal of shipping work that affects hundreds of millions (or billions) of end users.
- •High bar: “not everyone is fit to join”
- •Key traits: high agency, raw IQ, mission alignment
- •Impact pitch: code shipped affects massive consumer bases
- •Attracting talent even versus OpenAI/Anthropic through ownership and scope
Founder origin story and YC pivot journey: from edtech to fine-tuning to support
They recount meeting at IIT Kharagpur, applying to YC with an education idea, and pivoting under YC’s influence. Visa issues forced them to do YC remotely initially; later, they pivoted into fine-tuning and then found customer support through customer discovery.
- •Met at IIT Kharagpur; turned down prestigious offers to start a company
- •Applied to YC with an education/jobs platform; interviewed by Harj
- •Visa denials forced a remote YC experience, making execution harder
- •Pivot sequence: education → fine-tuning/inference → customer support platform
Why fine-tuning-as-a-service didn’t work: competing against the model curve
Esha explains their earlier fine-tuning business had traction when frontier models were expensive, but the funnel broke as OpenAI/Anthropic released cheaper, better models. They concluded they were “betting against AI” and needed a strategy that benefits from model progress.
- •Customers came for cost optimization when GPT-4-class models were pricey
- •New model releases made alternatives cheaper/better and pulled customers away
- •Constant retraining/keeping up created structural churn pressure
- •Decision: stop betting against model progress; build on top of it
The future: context-rich ops automation platform beyond support (plus founder advice)
They outline a broader vision: customer support is the entry point to gather context, data, and operational footing inside enterprises. Over time, Giga aims to expand into other OpEx-heavy functions (like compliance) and build a platform that enables broader enterprise optimization.
- •Long-term bet: context + enterprise data wins, not just smarter base models
- •Support provides deep customer interaction context to power other workflows
- •Expansion targets: other OpEx-heavy functions (e.g., compliance)
- •Vision: consolidate workflows into an enterprise ops automation platform
