Jellyfish

Executive Summary

Jellyfish is an engineering management platform (EMP) founded in 2016 by Andrew Lau, Philip Braden, and David Gourley — three Endeca veterans who scaled that company from 8 to 550 employees before its $1.1B sale to Oracle.^[1] The platform's original pitch was giving engineering leaders the visibility Salesforce gave sales: allocation, delivery, and cost data mapped to business objectives.^[1] Its AI Impact module, expanded into an end-to-end SDLC view in August 2025, applies that machinery to the question every engineering leader now faces — which AI coding tools are worth the spend.^[2]

AI Impact measures three dimensions — adoption and usage, spend by tool/team/initiative, and delivery outcomes (throughput, cycle time, quality) — across coding assistants (GitHub Copilot, Cursor, Claude Code, Amazon Q, Gemini Code Assist, Windsurf, Sourcegraph), agentic tools (Devin, Google Jules, Copilot Agent), and AI code-review agents (CodeRabbit, Graphite, Greptile).^[3][2] The benchmark behind it draws on 20M+ analyzed pull requests, and the company serves 500+ customers including DraftKings, Keller Williams, Blue Yonder, TaskRabbit, and Hootsuite.^[3][2]

---

Product Overview

Jellyfish connects to the systems engineering teams already use — Git, planning tools (Jira, Linear, Azure Boards), and AI tool usage/spend data — and derives intelligence from that data automatically, with no tagging, manual setup, or process changes required.^[3] AI Impact sits on top as a vendor-agnostic framework for comparing AI tools on consistent measures, explicitly positioned against "vendor-reported activity metrics" like a tool's own acceptance-rate dashboards.^[3] The August 2025 release extended coverage from coding assistants to agents and AI code review, with multi-tool comparison and usage-based spend tied to outcomes at team, individual, and initiative levels.^[2]

Key Capabilities

Product Surfaces

---

Technical Architecture

Jellyfish is a managed, closed-source SaaS platform. Attribution deserves precision: AI Impact does not run model-based detection to classify whether code was AI-written. It derives signals from existing engineering data — Git activity, planning systems, and AI tool usage and spend data — detected automatically without heavy integrations or manual tagging.^[3] That makes rollout light (no workflow changes) but means AI attribution is correlational, inferred from tool seats, usage telemetry, and delivery data rather than per-line provenance.

Key Technical Details

---

Strengths

• Broadest AI tool coverage in the category — assistants, agentic tools, and AI code-review agents in one vendor-agnostic comparison framework; coverage grew from Copilot-only to multi-tool as teams went multi-vendor^[2][5]
• Benchmark depth — AI Impact analysis draws on 20M+ pull requests, and Jellyfish's research found 90% of teams using AI coding tools in 2025, up from 61% the prior year — data scale smaller entrants cannot match^[3][2]
• Spend is a first-class metric — usage-based AI spend, including token costs, tied to outcomes at team/individual/initiative level; most competitors measure adoption but not the bill^[2][4]
• Low-friction attribution — signals derived from existing Git and planning data with no tagging or process changes required^[3]
• Established vendor — founded 2016, $114M+ raised from Accel/Insight/Tiger Global, 500+ customers, and a repeat founding team with a $1.1B prior exit^[1][2]
• CFO-grade surface — DevFinOps (capitalization, R&D tax credits, SOC 1 Type II) makes the same data audit-ready, which pure AI-measurement tools do not offer^[4]

---

Cautions

• Correlational attribution — deriving AI impact from existing delivery data avoids invasive instrumentation, but it infers rather than proves causation; teams expecting per-line AI provenance will not get it^[3]
• Numbers without prescriptions — a recurring critique is that Jellyfish surfaces metrics but does not always tell you what to do with them, and metrics disconnected from operational standards risk becoming performance surveillance rather than improvement levers^[6]
• Setup complexity — G2 reviewers cite complex initial configuration and data mapping, a steep learning curve for new managers, unclear documentation, and clunky team-hierarchy management^[7]
• Limited customization — reviewers report the lack of report customization limits adapting outputs to specific needs^[7]
• No public pricing, no self-serve — seat-and-module quotes through enterprise sales only; evaluation requires a demo cycle^[4]
• Module economics — AI Impact is one of three paid modules; buying AI measurement alone still means buying into an EMP vendor relationship^[4]

---

Pricing & Licensing

Pricing is based on number of seats and modules selected; there is no free trial or published price list.^[4]

Licensing model: Proprietary closed-source SaaS, annual enterprise contracts quoted per seat and module.^[4]

Hidden costs: Implementation and data-mapping effort during onboarding (a consistent G2 complaint), plus admin time to maintain team hierarchies and configurations as the org changes.^[7]

---

Competitive Positioning

Direct Competitors

When to Choose Jellyfish Over Alternatives

• Choose Jellyfish when: you run multiple AI coding tools and agents and need one consistent framework for adoption, spend, and outcomes — especially if finance also wants capitalization and audit-ready reporting^[2][4]
• Choose DX when: developer experience and survey-grounded research are the primary lens
• Choose Swarmia when: you want transparent pricing and a team-level, developer-trusted tool without an enterprise sales cycle
• Choose LinearB when: you want metrics paired with workflow automation acting directly in the PR pipeline

---

Ideal Customer Profile

Best fit:

• Mid-to-large engineering orgs (hundreds of contributors) running several AI coding tools simultaneously and needing to decide which to expand or retire^[2]
• VPs of Engineering who answer to CEOs/CFOs on AI ROI and want defensible numbers rather than vendor-reported metrics^[3]
• Companies that also need software capitalization, R&D tax credit, and audit-ready reporting from the same data^[4]

Poor fit:

• Small teams wanting self-serve, transparently priced analytics^[4]
• Teams expecting per-line AI code provenance or model-based detection of AI-written code^[3]
• Engineering cultures allergic to top-down metrics — without operational grounding the dashboards can read as surveillance^[6]

---

Viability Assessment

Jellyfish entered the AI measurement race with a structural advantage: the delivery, allocation, and spend plumbing was already built, so AI Impact is a lens over data it was already collecting. The risk profile is the inverse — a four-year-old funding round, an enterprise-only motion, and competitors (DX, Swarmia, LinearB) all converging on the same AI ROI question from cheaper or more developer-friendly angles.

---

Bottom Line

Jellyfish is the incumbent-scale answer to "is our AI tooling spend working" — the widest tool and agent coverage in engineering intelligence, spend tracked next to outcomes, and a 20M+ PR benchmark, all riding on an established EMP with 500+ customers.^[3][2] The trade-offs are the EMP heritage itself: enterprise-quoted modules, real onboarding effort, correlational rather than provable attribution, and dashboards that tell you what happened more readily than what to do next.^[7][6]

Recommended for: Mid-to-large engineering orgs standardizing measurement across multiple AI assistants and agents, especially where leadership needs board- and CFO-grade AI ROI reporting.

Not recommended for: Small teams wanting self-serve transparent pricing, buyers needing per-line AI provenance, or developer-led cultures that will reject top-down metrics.

Outlook: Positive. The August 2025 end-to-end SDLC release — agents and AI code review included — shows Jellyfish moving as fast as the tool landscape it measures; the open questions are fresh capital (nothing disclosed since the 2022 Series C) and whether the EMP bundle holds up against lighter, cheaper AI-measurement entrants.^[2][1]

---

Research by Ry Walker Research • methodology