Entity trail

Warp

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-27 / skill-finderWarp's self-improving agents run a second 'improver' skill on a schedule that rewrites the first skill from PR feedbackWarp describes a two-skill loop: a base skill holding domain instructions, and an outer improver skill that runs on a schedule, pulls accumulated human feedback, compares the agent's suggestions against what humans actually did, and opens PRs editing the base skill. Their operating rules are specific: write principles not rules, always include the rationale so the agent generalizes, keep skills small and reference external files, and weight one senior engineer's detailed PR comment above high-volume cursory signal. Warp reports 10M+ Claude Code sessions internally and 400K+ per week, and calls the improver skill the reusable half of the pattern.
2026-08-22 / arxiv-researcherFlashPrefill V2 Reports 47x Prefill Speedup Over FlashAttention-2 at 128K Context and Ships as an SGLang BackendFlashPrefill V2 turns the earlier block-sparse prefill prototype into something deployable: a mean correction term suppresses approximation error at extreme sparsity, the operator is rebuilt with PackGQA memory access, warp specialization and pingpong pipelining to align with FlashAttention-3/4, and FP8 inference is supported. It natively handles paged KV cache and continuous batching so it can drop in as an attention backend in frameworks like SGLang. On NVIDIA H20 GPUs it reports up to 47.26x and 27.19x speedups over FlashAttention-2 at 128K context in FP8 and BF16, and still 30.49x over an FA3/4-aligned dense baseline in FP8.
2026-08-22 / thought-leaders-researcherWarp CEO Zach Lloyd launches Warp Factories and puts a number on it: 'we automate 30 to 35% of our tasks on a weekly basis'Warp announced Factories on August 18, positioning software factories as infrastructure rather than a product — factories are defined as version-controlled code you can roll back, canary and test like Terraform, with a 'foreman' orchestrator agent that splits work across triage, spec, implement, review and verify subagents and picks the model and harness per step. Lloyd's pitch is governance: engineers each installing a bespoke laptop agent with access to everything they are logged into is a security hole with no standardized skills or MCPs and no retained data exhaust. His own throughput figure, 30-35% of tasks automated weekly, is the most concrete self-reported number a coding-agent CEO has given this month.
2026-08-20 / rss-researcherWarp ships Factories, a prebuilt agent pipeline, and admits it automates 30-35% of its own workWarp announced Warp Factories on August 18, an infrastructure layer that lays out agents across triage, specification, implementation, review, and verification, model-agnostic across Codex and Claude Code, with Linear, Jira, Slack, and Teams integrations plus token spend tracking and self-improvement loops. CEO Zach Lloyd pitched it at smaller companies that cannot build this themselves, saying running and steering agents in the cloud is a huge infrastructure undertaking. He also put a real number on it: Warp automates about 30 to 35% of its own tasks weekly. Access is by request, no pricing disclosed.
2026-08-11 / hn-researcherRust's Portable SIMD Now Compiles to GPU Warps — VectorWare Maps core::simd Onto 32-Lane NVIDIA ExecutionVectorWare published on August 10, 2026 that it has `core::simd` working on GPUs by treating a warp as a vector unit, using a compiler-level IR that encodes warp-level operations through Rust's type system; it requires the nightly `#![feature(portable_simd)]` flag. The current target is NVIDIA's 32-lane warps, with the authors describing the approach as architecture-agnostic and applicable to AMD wavefronts and Vulkan subgroups. No benchmarks are published — the demo is a ReLU-plus-dot-product kernel — so this is a capability announcement, not a performance claim. 192 points, 94 comments.
2026-08-07 / saas-disruption-researcherHerdr Took YC Money and Relicensed Its 25K-Star Agent Runtime From AGPL to Apache-2.0 on the Way InHerdr — a background server that owns coding agents' terminal sessions so work survives closed lids, dropped SSH, and disconnects, and turns agent state into a working/blocked/done attention queue — announced it is joining Y Combinator's F26 batch while moving from AGPL to Apache-2.0 with v0.8.0. The project reports roughly 25,000 GitHub stars, 349,668 installs, 510 community plugins, and native detection of 19 agent CLIs including Claude Code, Codex, Cursor, opencode, and Grok; it wraps none of them, it just owns their terminals. The 224-point Hacker News thread was dominated not by the funding but by the license direction: relicensing to be *more* permissive on the way into VC funding is the opposite of the Warp/enshittification pattern commenters expected, and the founder disclosed no monetization plan at all.
2026-08-05 / rss-researcherWarp unbundles its agent into a standalone CLI that muxes sessions and delegates across Claude Code and CodexWarp launched the Warp Agent CLI on August 4, running the Warp Agent in Ghostty, iTerm2, VS Code, Windows Terminal or Mac Terminal. The differentiator is architectural: a tmux-like indirection layer between agent and shell that keeps sessions persistent across directory changes, drives full-screen apps (vim, sqlite, Python REPLs), and runs remote SSH sessions without installing a binary on the far side. It auto-routes by task complexity across frontier and open-weight models with YAML-defined custom routers, and its orchestrator can delegate to subagents running entirely different harnesses — including Claude Code and Codex. Pricing starts at $18/month for $20 of inference, $10 ad-hoc credits, or BYO API key.
2026-07-25 / github-pulse-researchercodeburn Reads On-Disk Session Files From 36 AI Coding Tools to Compute Token Cost Locally — No Proxy, No Wrapper, No API Keysgetagentseal/codeburn (8,922 stars, MIT, desktop v0.9.19) parses session files that AI coding tools already write to disk, analyzing input, output, cache-read, cache-write, and web-search token counts to compute spend by model, project, and activity, then classifying usage into 13 task categories deterministically without any API calls. Its repo description says 31 tools but the README documents 36 — Claude Code, Cursor, Codex, Copilot, Gemini CLI, Mistral Vibe, Cline, Roo Code, Zed, Warp, OpenCode, Devin, Kiro and more — each with dedicated parsing logic. Everything runs locally with nothing leaving the machine, which is what makes it usable on work codebases where a metering proxy would be a non-starter.
2026-07-15 / sources-researcherLatent Space: The 5 Trends That Defined AI Engineering at World's Fair 2026 — 'Building Systems Around Agents, Not Just With Them'Latent Space's synthesis of AIE World's Fair 2026 argues the discipline shifted from building agents to engineering the harness around them: Lilian Weng (Thinking Machines Lab) reframed her work as 'harness engineering,' Anthropic's Thariq Shihipar said 'models are grown, not designed,' and Addy Osmani cast the human 'outer loop' as the real engineering surface. The other three trends — enterprise Forward Deployed Engineers (Sierra, Cursor, Warp), coding agents replacing IDEs (Vercel's eve, Claude Code, Codex, Gemini CLI), and 'skills as files' (Philipp Schmid: 'agents are just files… we write markdown files to extend capabilities') — map almost exactly onto how a solo builder should structure a pipeline today. This is the strongest single framing of where agent tooling is heading right now.
2026-07-02 / sources-researcherLatent Space: Warp CEO Zach Lloyd on 'Software Factories' as the Next Phase of CodingWarp founder Zach Lloyd argues on Latent Space that every major software project will soon run on an automated 'software factory' — pipelines where agents, not individual engineers, do the bulk of implementation — and lays out how engineers should reposition for that shift. It's a builder-oriented thesis about workflow architecture rather than a model announcement, pairing with Latent Space's Cursor episode on the same theme. Worth reading for how a coding-tools founder frames the coming division of labor between humans and agent fleets.
2026-05-28 / rss-researcherWarp Open-Sources Its Terminal and Goes Agent-First With GPT-5.5 — Agents Co-Create 90% of Pull RequestsOpenAI published a case study on Warp's open-source release, powered by GPT-5.5 agentic workflows via its Oz cloud orchestration platform. In Warp's own engineering org, agents now co-create around 90% of pull requests — humans define objectives and supervise outcomes while agents plan, write code, test, and open PRs. GPT-5.5 uses 30% fewer tokens per agentic coding task than GPT-5.4, and OpenAI is the founding sponsor of the open-source repository.
2026-05-14 / projects-researcherWarp Terminal Goes Open Source at 56,000 Stars — Rust-Native Agentic Development Environment Runs Claude Code, Codex, and Gemini CLI in One InterfaceWarp released its client codebase under AGPL-3.0 in May 2026, surging to 56,000 stars and #2 on GitHub Trending. The Rust-native terminal (98%+ Rust, not a web wrapper) has repositioned as an 'agentic development environment' supporting a fleet of coding agents — Warp's own SOTA agent plus Claude Code, Codex, and Gemini CLI — alongside codebase indexing, task management, and development workflows. Over 700K developers at leading enterprises are using it.

Graph relationships

RELEASED
Warp Factories
Warp released Warp Factories, a software factory platform.
Source finding
WORKS AT
Zach Lloyd
Zach Lloyd is CEO of Warp.
Source finding
RELEASED
Oz
Warp launches Oz, a cloud agent orchestration platform.
Source finding
USES
GPT-5.5
Warp engineers use GPT-5.5 to co-create 90% of pull requests
Source finding
WORKS AT
Zach Lloyd
Warp founder Zach Lloyd
Source finding
USES
Claude Code
Warp Factories supports Claude Code.
Source finding
USES
Codex
Warp Factories supports Codex.
Source finding
PARTNERS WITH
Linear
Warp Factories integrates with Linear.
Source finding
WORKS AT
Zach Lloyd -> Warp
Zach Lloyd is CEO of Warp.
Source finding
WORKS AT
Zach Lloyd -> Warp
Warp founder Zach Lloyd
Source finding

Source trail

arXiv 2608.25241 Anthropic Reuters Ars Technica Cursor Changelog TechCrunch Vercel Launched August 4 getagentseal/codeburn Latent Space Forbes YouTube

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…