Continue the conversation — chat opens pre-seeded with the current signal, caps, and movement.
AI code review platform for GitHub-native engineering teams. Cubic 2.0 (January 2026) rebuilt the detection engine with AI Wiki, codebase caching, and a learning loop — vendor reports actionable-comment rate moved 20% → 60%+.
Target user: GitHub-native engineering teams of seed to growth-stage companies that want depth and precision in code review rather than breadth and scale. Independent comparisons consistently position Cubic as the depth/precision specialist vs CodeRabbit's breadth/scale.
Differentiator: Cubic 2.0 detection engine (AI Wiki + learning loop); enterprise-tier SAML/SSO and GitHub Enterprise support; customer roster across credible engineering organizations (Cal.com, n8n, Resend, Better Auth, Granola, Legora, Browser Use, Cartography); MCP/Skills (Launch Week 03) enables inbound connectivity from Claude Code and Cursor.
Adoption & Proof Points
- **Public customer roster:** Cal.com, n8n, PostHog, Resend, Better Auth, Granola, Legora, Browser Use, Cartography (plus Linux Foundation projects referenced in third-party coverage)
- **Published case study:** Browser Use — PR cycle time days→~3 hours, 85% faster merges, halved tech debt (Tesla/Netflix/IBM/Adobe/NVIDIA are Browser Use's customers, not cubic's)
- **Investor network:** Vercel Ventures, PeakXV, Jared Palmer, Theo Browne, Zeno Rocha; founders Paul Sanglé-Ferrière & Allis Yao
- **YC batch:** YC profile/LinkedIn list X25 (the internal record carries S25 — minor discrepancy, same company)
- **Active product cadence:** Cubic 2.0 (Jan 12, 2026), Launch Week 03, Jira/Linear/Asana integrations now GA, steady April–May blog cadence
- **Independent documentation:** ZenML LLMOps Database case study on cubic's false-positive architecture (51% FP reduction, ~11% FP rate)
- **Benchmark caveat:** Cubic markets #1 on Martian Code Review Bench (61.8% F1, March 2026) — but the claim is contested. Qodo (64.3% / hardest bugs), CodeRabbit (51.2% online), CodeAnt, and Kilo-Code have each published competing #1 claims with different methodologies. Martian itself is an independent benchmark, so the capability signal is third-party-validated; only the '#1' superlative is disputed
Risks & Limitations
- **Contested benchmark headline:** the '#1 Martian' marketing claim is disputed by Qodo, CodeRabbit, CodeAnt, and Kilo-Code with different methodologies; treat as `[vendor-only, contested]`
- **Team size:** ~3 people per YC profile; no Series A announced; no public ARR or customer-count
- **Platform constraint:** GitHub-only — no GitLab, Bitbucket, or Azure DevOps support
- **Governance gaps:** no published audit logs, no RBAC, no self-hosted option, cloud-only, no data residency, SOC 2 Type 2 not confirmed (Type 1 only)
- **Cap-free but not Assessed-eligible:** advancement requires GitLab/Bitbucket support, audit logs/RBAC, SOC 2 Type 2, and Series A funding
Capabilities & Integration
**Autonomy (10, Developing/Basic boundary):** background agents create fix PRs autonomously; codebase scanning runs nightly; Cubic 2.0 (Jan 12, 2026) added AI Wiki and learning loop. Humans still execute fixes — the 20% → 60% actionable-comment rate is a precision win, not an autonomy band change.
**Integration (13, Capable): ** core GitHub PR integration; MCP/Skills (Launch Week 03 Day 4) enable inbound from Claude Code, Cursor, VS Code, Codex, and Gemini CLI. Jira/Linear/Asana integrations now GA with acceptance-criteria checking and inline ticket context; Confluence at Pro; Slack at Pro. GitHub-only for VCS (peers offer GitLab); no CI/CD pipeline-native depth.
**Context (13, Repository-Deep):** Cubic 2.0 added AI Wiki (pre-mapped codebase context) and live documentation integration. Codebase caching/sandbox snapshotting for performance. 150-file per-PR ceiling remains a practical limit.
**Interface (10, Developing): ** GitHub PR comment surface (primary); CLI for local pre-commit reviews; MCP/Skills inbound connectivity; Coding Analytics (April 17). No dedicated VS Code/JetBrains extension.