Zach Lloyd

Zach Lloyd is the founder and CEO of Warp, where he has led its evolution from a modern terminal into infrastructure for coordinating coding agents. His argument is that useful automation requires engineers to design the context, permissions, feedback and review processes around agents. In a July 2026 interview, he described this work as “building the system that builds the product.”[1][8]

Before that agent-focused work, Lloyd helped build collaborative productivity software at Google and co-founded SelfMade. Those experiences supplied concrete context for Warp: the engineering difficulty of modernizing familiar tools, and his decision to build a company whose product was software itself.[2][3]

Lloyd led engineering on Google Sheets and subsequently the Google Docs suite. In his December 2022 Jamstack Radio interview, he explained the spreadsheet’s central tradeoff: the web platform made it inherently collaborative and accessible through a link, while making Excel-like performance difficult. He also recalled the tension between improving document editing and supporting users’ existing Microsoft Office files.[2][3]

Lloyd co-founded SelfMade in 2016 and stepped down as CTO in 2019. In Warp investor Sequoia’s profile, he recalled realizing that social media was not a problem he personally cared about, and that SelfMade’s product depended on people’s labor. He wanted to work where software was the product. He then turned to improving the terminal; Sequoia dates Warp’s founding to June 2020.[3]

Lloyd recruited former Google colleague Shikhiu Ing as Warp’s founding designer. Their prototype helped communicate the proposed experience, and engineers Michelle Lim and Aloke Desai joined the early team. Lloyd recalled that the technical challenge was to innovate while keeping users’ existing scripts and muscle memory working.[3]

Warp changed two basic terminal interactions: editing commands and handling their output. Its input editor supported familiar mouse-based editing, while output blocks kept a command and its results together as a unit that could be shared. Lloyd said an Electron prototype consumed too many resources and performed poorly when rendering large amounts of output. The team chose Rust, learning it with help from an early teammate who already had experience, and built its own UI framework and rendering stack, including shaders and direct access to Apple’s Metal graphics API. He considered the additional engineering overhead worthwhile for performance and control over the application.[2]

On June 24, 2025, Lloyd introduced Warp 2.0 as an “Agentic Development Environment,” combining coding, agent management, terminal interaction and shared context. A universal input accepted prompts and commands; Warp Drive stored reusable commands, notebooks, prompts, rules and MCP configuration for teammates and agents. His rationale was that agents needed a team’s accumulated knowledge and conventions alongside access to its code.[4]

Lloyd treated autonomy as separate decisions. Warp 2.0 provided distinct permissions for reading files, executing commands and accepting code changes, plus command allowlists and denylists and the ability to pause agents. He defended starting with local execution because, given the technology’s capabilities at the time, he believed cloud agents would remove humans from the loop too early. He also argued that developers should be able to adopt agents gradually, retaining the option to use Warp solely as a terminal.[4]

During Warp 2.0’s development, Lloyd required engineers to start every coding task with a prompt in Warp. His July 2025 account paired that mandate with practical fallbacks: share successful prompts internally, report unproductive prompting after ten minutes, try another coding agent, and code manually when that was faster. The policy made the team’s everyday work a way to test the product and expose failures.[5]

Engineers remained responsible for understanding submitted code and meeting the same quality standards as handwritten work. Lloyd’s rule was “AI is a tool, not a responsible party on our software team.” He advised specifying architecture, APIs and tests; discussing plans before implementation; and making small, verifiable changes rather than generating a complex feature in one attempt. Persistent rules and saved prompts captured repeated conventions and workflows.[5]

Warp launched Oz on February 10, 2026, extending agent workflows into the cloud. Oz used Docker-based sandbox environments, shared team context and access-controlled run links with audit trails. It supported changes across repositories and interactive, programmatic or scheduled runs independently of Warp Terminal. Lloyd’s emphasis moved toward centralized oversight of agent work rather than supervision confined to an individual laptop.[1]

On April 28, 2026, Lloyd announced the release of the Warp client under AGPL, with Oz-assisted community contribution workflows and OpenAI as the repository’s founding sponsor. He argued that agents had shifted the development bottleneck toward specification and behavioral verification: contributors could shape features and check results while agents performed implementation work. He also explicitly connected the decision to competition and limited resources, presenting openness as a way to accelerate development.[6]

Warp’s contribution process gives that division of labor a concrete form. Issues are triaged before implementation; feature work requires an approved specification, while actionable bug fixes can proceed directly to code. Specifications pair testable behavior invariants with a technical plan and validation methods. Pull requests receive Oz review followed by human subject-matter-expert review, then CI and merge. Contributors must also provide proof of manual testing.[7]

In his July 2026 interview, Lloyd described progressing from one-off cloud-agent runs to timers and then to automating the main development loop: triage, specification, implementation, review, verification, shipping and monitoring. Teams would choose repositories, the stages to automate and the points requiring human involvement. He envisioned this working through existing issue trackers, communication tools and code repositories, with engineers improving agents’ context, feedback and workflows. Warp’s own open-source client served as a test bed for that approach.[8]

Lloyd announced Warp Factories in closed beta on August 18, 2026. He chose to offer infrastructure that organizations could build on and customize. Factory definitions were version-controlled code, supporting testing, rollback and gradual deployment. The design allowed different models and coding harnesses, evaluations on a team’s own tasks, and observer agents that could propose changes to factory configurations. These mechanisms made the automation process itself something engineers could revise and verify.[9]

Lloyd expected adoption to begin with specific use cases and lower-risk work, while difficult, ambiguous and greenfield tasks continued to require people. In July 2026, he predicted that every significant software project would have something resembling a factory within the following year. His August announcement predicted CI/CD-like ubiquity “in the next few years.” Both were forecasts; his practical advice was to begin by automating an annoying part of the job and discover the bottlenecks by trying to build the loop.[8][9]