The agentic development environment born out of the terminal
HQ: New York, NY, United States (offices in New York and San Francisco) | Founded: 2020 | Employees: ~100 (102 on LinkedIn, mid-2026, up 43.7% YoY) | Stage: Series B; ~$73.6M total raised, valuation not disclosed | Website: https://www.warp.dev
Warp was founded in June 2020 by Zach Lloyd, a former Google engineer, on a specific grievance: every terminal emulator on the market was a decades-old design that developers tolerated rather than chose. Lloyd's pitch to GV, made in early 2020, was to rebuild the terminal from scratch. The first check came within months, and the company spent its first year proving it could ship a fast terminal at all — initially as an Electron prototype, then rewritten natively in Rust for performance, with a custom GPU-rendered user interface built with help from Atom's creator.
It launched publicly in April 2022 with a Hacker News post that produced more than 10,000 signups in 24 hours, alongside a $17 million Series A led by Dylan Field, Figma's founder and CEO — an unusual entrepreneur-led round that also included Marc Benioff through TIME Ventures, Jeff Weiner through Next Play Ventures and Elad Gil. A $50 million Series B from Sequoia followed in June 2023, with Andrew Reed joining the board and Sam Altman and Tobi Lütke joining as angels. That is where the funding record stops: no disclosed primary or secondary round since.
The AI pivot came in stages. A language-model sidebar shipped in 2023, before most competitors. Agent Mode, launched in June 2024, gave the model direct terminal access for multi-step work — Warp claims it was first to use the name. Linux and Windows builds followed, then a GPU rendering engine rewrite in March 2026, then the biggest strategic decision of the company's life: in April 2026 Warp open-sourced its client under AGPL-3.0, arguing it was competing against better-funded closed-source rivals and needed a different distribution argument. The repository passed 40,000 stars within a day and now sits near 65,000.
The 2026 product line has three parts. Warp itself is the agentic development environment, ranked first on Terminal-Bench at launch in June 2025 and in the top tier of SWE-bench Verified. Oz, launched in February 2026, orchestrates hundreds of coding agents in parallel in Docker sandboxes with audit trails, a scheduler, human steering and self-hosting. Warp Factories, launched in August 2026, is the most ambitious piece: infrastructure for a company's own software factory, defined in version-controlled configuration, with a foreman agent delegating to triage, spec, implementation, review and verification subagents, scored by LLM judges.
The company's internal numbers are the strongest evidence it is real. Oz was writing 60% of Warp's pull requests at launch; agents now co-create around 90%. Users launch roughly three million agents a day, and Warp reports 250 million lines of code generated per week with a 97% acceptance rate on agent-suggested diffs. A "walk" factory automates about 75% of changes to Warp's own marketing site, while a "run" factory is the ambition.
The uncomfortable part is the money. Warp buys inference at API rates and sells it at roughly a 20% markup, and the CEO has said plainly that competing on model cost is a losing battle for a startup. So the thesis has to be that the value is in orchestration, context and governance rather than in the tokens — a claim the fastest-growing part of the business, Oz and Factories, is built to test.
Warp started as a nicer terminal and has become something more interesting: a control plane for software being written by agents. The company has repositioned three times in six years — terminal, AI terminal, agentic development environment, cloud orchestration, software factory — and each pivot has been defensible because of the one asset that never changed. The terminal sits at the bottom of the developer stack and sees everything: every command, every output, every error. An IDE fork sees less.
The bet is that context is the constraint, not model quality. Warp's agents are explicitly model-agnostic — Claude, GPT, Gemini, GLM, Kimi, MiniMax, Qwen — and Warp will orchestrate Claude Code and Codex as happily as its own agent, because the CEO argues that engineering leaders are finished choosing a favourite coding agent. Oz runs hundreds of agents in Docker sandboxes with audit trails and self-hosting. Warp Factories, launched in August 2026, is factory-as-code: version-controlled definitions of a triage, spec, implement, review and verify pipeline, with LLM-as-a-judge scorers.
Two things give this unusual shape. The client is AGPL open source since April 2026, with OpenAI as founding sponsor and roughly 65,000 GitHub stars — a deliberate move against better-funded closed-source rivals. And roughly 90% of Warp's own pull requests are now co-created by agents, so the product is its own largest reference customer.
The open questions are unglamorous. Warp has raised nothing since June 2023 while competing against companies with billions behind them. Inference is sold at API rates plus a 20% markup, which is a thin business by design. And the terminal itself is free and excellent — Ghostty alone takes about a million macOS downloads a week — so Warp's own answer was to ship its agent into its competitors' terminals as a standalone CLI. That is a clever hedge, and it is also an admission that the terminal was never the business.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki