Prasanna Sankar

Prasanna Sankar is co-founder of Vorflux, which builds cloud-based agents that plan, implement, test, and review software changes. A former co-founder and chief technology officer of Rippling, Prasanna is working to reduce the supervision that coding agents require between receiving an engineering request and delivering a change for review. His approach centers on giving agents a running application, engineering-team practices, and feedback from tests and other models.

Building connected employee software

Prasanna studied computer science at the National Institute of Technology, Tiruchirappalli. His early career included an internship at Google, software development at Microsoft, and co-founding LikeALittle, a Y Combinator Winter 2011 company. He later became director of engineering at Zenefits and the first engineer on its payroll product, working in employee software alongside Parker Conrad.

In 2016, Prasanna and Parker co-founded Rippling. Prasanna had approached Parker about starting another company after Parker left Zenefits. Working initially from Parker’s home, they built a product connecting payroll and benefits with application access, email groups, and other workplace systems. An employee’s arrival or departure required coordinated changes across those tools; Rippling brought those actions together.

As Rippling grew, Prasanna’s responsibilities shifted toward aligning an engineering organization of roughly 100 people. He described struggling with the meetings that work required: he avoided meetings he did not enjoy, recognized that this made him less effective, and handed engineering leadership to someone more enthusiastic about the role. After stepping back as CTO, he became increasingly interested in blockchain’s technical possibilities.

A unified interface for crypto

With 0xPPL, Prasanna pursued a unified interface for crypto activity. In announcing its 2.0 release, he explicitly connected its ambition to Rippling: simplify a fragmented collection of systems through one product. The application combined a wallet, portfolio tracking, and tools for moving and managing assets across blockchains.

The engineering problem extended beyond displaying wallet balances. Funds could sit inside smart contracts, each requiring particular interactions to withdraw or unstake them. Prasanna’s team used LLM-generated adapters to connect those contracts to the interface. He explained that open-source contract code and the ability to simulate interactions made generated adapters practical. A shared gas tank also reduced the need for users to manage transaction fees separately on each chain.

Prasanna argued for investing in that infrastructure rather than chasing short-lived crypto trends: as chains and applications multiplied, a universal interface would become more useful, while code generation made it more feasible. The ambition did not translate into the adoption the team needed. 0xPPL wound down in June 2026, and Prasanna acknowledged that its crypto social network had not worked. Vorflux followed, applying AI-generated software to the engineering workflow itself.

Giving agents the environment to finish

Prasanna introduced Vorflux with $15 million in funding. His premise is that stronger code generation leaves substantial work with the developer: preparing environments, clarifying plans, coordinating changes, testing behavior, and reviewing results. An agent that cannot run the application sends those tasks back to a person’s laptop.

Vorflux’s approach makes three engineering choices explicit:

• Run the whole application. Vorflux provisions cloud machines where agents can work with the frontend, backend, databases, and browser together. They can execute the application, run tests, and investigate failures before returning a pull request. Its session tools support parallel work and include testing artifacts, screenshots, and recordings so reviewers can inspect behavior alongside code.
• Use different models for different jobs. The team builds the orchestration around existing models rather than training a foundation model. Planning, implementation, and review can use different models, including a reviewer from another provider. This lets the company change which model handles a task as capabilities change.
• Turn corrections into reusable engineering guidance. Prasanna argues that a missed requirement should prompt an investigation into why the agent could not find or apply it. His examples include adding tests, code comments, reviewer instructions, skills, or memory. The aim is to make the lesson available to later sessions rather than repeatedly correcting the same mistake.

Prasanna calls the intended workflow supervision by exception: people set direction and review consequential results while agents perform more of the intervening work. In his launch account, he described giving Vorflux two links for a feature requested by Giga and receiving an 8,000-line pull request by morning, with testing and a video of the result. That is his reported example of the workflow, rather than a general measure of its performance.

His position on engineering bottlenecks follows from this approach: after each model improvement, identify where people still spend time and use more computation to address it. He argues that teams should exhaust what agents can do before hiring more engineers, and that faster implementation shifts attention toward deciding what to build and whom to sell it to. Vorflux puts that position into a concrete product requirement: an agent should return an executable, tested change with enough evidence for a person to review.