← All speakers

Bio, Work & Ideas

Sunny Rekhi

Conference affiliation: FDE CTO · Decagon · 2026

Sunny Rekhi is chief technology officer of forward-deployed engineering at Decagon and co-founded Tarsal, a cybersecurity infrastructure company acquired by Monad. His work focuses on converting complex enterprise integrations and customer-specific AI deployments into reusable products.

Before founding Tarsal with Manny Gundampalli, Rekhi worked on infrastructure, backend engineering, and data systems at TikTok and Uber, including Uber’s JUMP division. Tarsal entered Y Combinator’s Summer 2021 batch, with Rekhi serving as chief executive. Its security data pipelines collected and standardized information from cloud services, identity providers, and security tools for analysis in systems including Snowflake, Amazon S3, and Databricks. Customers included Robinhood and Navan before Tarsal joined Monad.

At Decagon, Rekhi leads work configuring enterprise customer-service agents: shaping their communication style, connecting business systems, defining authorized actions, and determining when humans should intervene. His account of deploying enterprise agents advances several specific principles:

  • Forward-deployed engineering is product engineering. Enterprise requests should improve the shared platform, with customer-facing engineers meeting the same technical standards as core product teams.
  • Custom work should become self-service. Repeated bespoke CRM integrations exposed a missing product capability; replacing them with configurable tools let customers and specialized agent builders handle implementations previously requiring engineers.
  • Architectural restraint becomes more valuable as code gets cheaper. AI-generated patches can quickly satisfy individual customers while creating brittle, opaque systems. Reusable integrations and customer-controlled configuration make deployments easier to maintain.
  • Prove business value before expanding. Historical support data helps identify high-return automation opportunities and establish success metrics early. Existing customer-service integrations can subsequently enable proactive outreach, as in Decagon’s work with Hertz.

Rekhi has also warned about prompt injection in external documents, including research papers, contracts, and reports analyzed by AI systems. He hosts Before It Clicked, a podcast examining how founders navigate customer discovery, false starts, and product-market fit.

Read the topics behind these talks

1 conference talk

Key ideas

Scroll to read ↓

Decagon’s deployment model connects agent configuration, customer advice, and reusable product capabilities so each enterprise deployment improves the next.

  • Start with the support interaction
    1:02 ↗
  • Configure the agent and improve the product
    2:56 ↗
  • Split the generalist role as the organization grows
    6:03 ↗
  • Exercise restraint when custom code is easy
    7:58 ↗
  • Agree on success before implementation
    9:11 ↗
  • Let industry knowledge accumulate
    10:22 ↗
  • Make engineering work available through self-service
    11:17 ↗
  • Sequence the deployment around early value
    12:41 ↗
  • Use support history to choose what to automate
    13:44 ↗
  • Build a company that learns from every deployment
    15:03 ↗

References