Devin (Cognition) | Ry Walker Research

Key takeaways

Raised $1B+ Series D at a $26B valuation (May 2026), up from $10.2B in September 2025 — one of the fastest markups in AI coding
Run-rate revenue reached ~$492M with ARR more than doubled since the Windsurf acquisition; targeting $1B ARR in 2026
Pricing collapsed from the old $500/mo era: Free, Pro $20/mo, Max $200/mo, Teams $80/mo + $40/seat, Enterprise custom
Now genuinely multi-model: OpenAI, Claude, and Gemini frontier models plus proprietary SWE-1.6 across Devin and Windsurf surfaces
Devin 3.0 (2026) added dynamic re-planning, desktop-app QA via computer use, and Devin Desktop unifying cloud/CLI/editor agents

FAQ

What is Devin?

An autonomous AI software engineer that can plan, code, debug, test, and deploy — working independently or alongside human engineers across cloud, CLI, and the Devin Desktop editor.

How much does Devin cost?

Free tier; Pro at $20/month; Max at $200/month; Teams at $80/month plus $40 per developer seat; Enterprise custom. The original $500/month single-tier era ended with Devin 2.0 in April 2025.

What is Devin best at?

Well-defined tasks at scale: security vulnerability fixes, migrations, test generation, and brownfield feature work. Mercedes-Benz compressed an eight-month legacy modernization to eight days.

Who uses Devin?

Citi, Mercedes-Benz, Goldman Sachs, Dell, Santander, Nubank, Itaú, and the U.S. Army and Navy, plus thousands of other companies.

Is Devin model-agnostic?

Increasingly yes. Cognition calls itself "an agent lab" working with multiple foundation model providers — OpenAI, Claude, and Gemini frontier models are available alongside its proprietary SWE-1.6, though models are platform-mediated rather than bring-your-own-key.

How does Devin compare to Tembo?

Devin is a single autonomous agent inside Cognition's platform. Tembo orchestrates multiple agents and integrates with existing tools like Jira.

Executive Summary

Devin is an autonomous AI software engineer from Cognition, the first mover in the "AI software engineer" category. In May 2026 Cognition raised more than $1B at a $26B valuation — up from $10.2B just eight months earlier — with run-rate revenue around $492M and a customer list spanning Citi, Mercedes-Benz, Goldman Sachs, Dell, and the U.S. Army and Navy.^[1]^[2] After acquiring Windsurf in July 2025, Cognition has evolved from a proprietary single-agent product into a multi-surface, multi-model platform.

Attribute	Value
Company	Cognition
Founded	2023
Funding	$1B+ Series D at $26B valuation (May 2026); prior $400M at $10.2B (Sept 2025)^[2]^[3]
Revenue	~$492M run-rate, as of June 2026; targeting $1B ARR in 2026^[1]
Headquarters	San Francisco, CA

Product Overview

Cognition is an applied AI lab — now self-described as "an agent lab" — best known for Devin, billed as "the first AI software engineer."^[4] Launched in March 2024, Devin was the first autonomous coding agent to gain mainstream attention, setting a new state-of-the-art on SWE-bench at launch (13.86% resolution rate vs. 1.96% prior best).

Devin has merged hundreds of thousands of PRs across thousands of companies.^[5] Enterprise usage grew more than 10x in the first months of 2026, and ARR has more than doubled since the Windsurf acquisition. Cognition reports that 89% of code committed by its own engineers is committed by Devin.^[1]

Key Capabilities

Capability	Description
Autonomous Execution	Plan and execute multi-thousand-decision engineering tasks independently
Dynamic Re-planning	Devin 3.0 alters strategy on roadblocks without human intervention^[6]
Desktop QA	End-to-end testing of desktop apps via computer use, with recorded test sessions^[6]
Full Stack	Build and deploy applications end-to-end
Code Review	First-pass review automation via Devin Review

Product Surfaces / Editions

Surface	Description	Availability
Devin Agent	Core autonomous engineer with shell, editor, browser	GA
Devin Desktop	Code editor + agent coordination across desktop, cloud, CLI	GA (2026)^[1]
Windsurf IDE	Cognition's AI IDE with Codemaps	GA
DeepWiki	Auto-generated documentation for large codebases	GA
Devin Review	First-pass code review automation	GA

Technical Architecture

Devin operates in a sandboxed cloud environment with full development tooling. Users assign tasks via Slack, web UI, the Devin Desktop editor, or CLI, and Devin works autonomously — reporting progress, accepting feedback, and collaborating on design decisions as needed.

Key Technical Details

Aspect	Detail
Deployment	Cloud (sandboxed compute); VPC deployment on Enterprise
Model(s)	Proprietary SWE-1.6 plus OpenAI, Claude, and Gemini frontier models^[7]^[1]
Integrations	Slack, GitHub, web UI, CLI, Devin Desktop
Open Source	No (proprietary)

Cognition develops custom models optimized for coding agent workloads — SWE-1.6 is "the most used model in Windsurf" — but now explicitly positions itself as working with multiple foundation model providers so customers can optimize performance and cost.^[1]

Strengths

First mover — Established "AI software engineer" category and brand recognition
Enterprise traction — Citi, Mercedes-Benz, Goldman Sachs, Dell, Santander, Itaú, U.S. Army/Navy^[1]
Deep pockets — $1B+ Series D at $26B valuation enables sustained R&D^[2]
Proven at scale — Mercedes-Benz compressed an eight-month modernization to eight days; Itaú auto-fixes 70% of security vulnerabilities^[1]
Multi-model platform — Proprietary SWE-1.6 alongside OpenAI, Claude, and Gemini frontier models^[7]
Expanding product suite — Devin Desktop, Windsurf IDE, DeepWiki, Devin Review

Cautions

Opaque usage metering — Cloud-agent compute (historically "ACUs") is hard to predict; developers call the units "entirely too opaque/confusing/complicated"^[8]
Struggles with ambiguity — Needs clear requirements; weak on open-ended or architectural work
Platform-mediated models — Multi-model, but no true BYOK; you consume models through Cognition's stack
Integration sprawl risk — Devin, Windsurf, Desktop, Review, and DeepWiki are still converging post-acquisition
Limited soft skills — Can't manage stakeholders, mentor reports, or handle interpersonal work
Valuation pressure — $26B on ~$492M run-rate prices in flawless execution^[2]

What Developers Say

Devin generates abundant — and polarized — discussion among practitioners:^[8]

"Overall I've been really impressed with Devin. IMO it's the best tool for AI generating features." — servercobra, Hacker News

"The first of the many 'code review' LLM-bots that have come out that doesn't actively feel like 'slop'." — samyok, Hacker News

"ACU are entirely too opaque/confusing/complicated. The entire description of them is shrouded in mystery." — joshstrange, Hacker News

"Under no circumstances should we allow AI to be the final judge of whether something should be merged." — rednafi, Hacker News

The pattern: genuine praise for Devin Review and well-scoped feature work, persistent frustration with usage-pricing opacity, and skepticism about fully autonomous merging.

Pricing & Licensing

Current published tiers (live pricing page, June 2026):^[7]

Tier	Price	Includes
Free	$0	Light agent quota, limited models, unlimited inline edits/Tab completions
Pro	$20/mo	OpenAI/Claude/Gemini frontier models, free SWE-1.6, Devin Cloud agents
Max	$200/mo	Pro with significantly higher usage quotas
Teams	$80/mo + $40/seat	Collaboration, centralized billing, admin analytics, priority support
Enterprise	Custom	VPC deployment, SAML/OIDC SSO, dedicated account team

Licensing model: Subscription (usage-quota based; extra usage at API pricing)

Note: The original $500/month single-tier era ended with Devin 2.0 (April 2025), which cut entry pricing to $20/month. Heavy cloud-agent usage is still metered and can be unpredictable on complex tasks.^[8]

Competitive Positioning

Direct Competitors

Competitor	Differentiation
Tembo	Tembo is agent-agnostic with Jira/Bedrock BYOK; Devin is single-vendor platform
Cursor	Cursor is AI IDE-first; Devin pairs an IDE (Windsurf/Desktop) with autonomous cloud agents
Claude Code/Codex	Model-native CLI tools; Devin is a complete platform with managed compute
Factory AI	Both enterprise-focused; Factory has Droids for parallel execution

When to Choose Devin Over Alternatives

Choose Devin when: You need proven autonomous execution at scale with enterprise support and a unified IDE + agent platform
Choose Tembo when: You need agent-agnostic orchestration with Jira/BYOK integrations
Choose Cursor when: You want AI-assisted IDE workflows, not fully autonomous agents
Choose Factory AI when: You need parallel Droids with enterprise workflow integration

Ideal Customer Profile

Best fit:

Enterprises doing large-scale migrations (legacy modernization, framework upgrades)
Organizations with security vulnerability backlogs (SonarQube, Veracode queues)
Companies needing test coverage improvements at scale
Teams wanting one vendor for IDE, cloud agents, code review, and docs
Clear, repetitive workloads with well-defined requirements

Poor fit:

Teams with ambiguous or rapidly changing requirements
Organizations requiring true BYOK or self-hosted models
Companies wanting deep Jira/enterprise workflow orchestration across multiple agents
Teams sensitive to unpredictable usage-based costs

Viability Assessment

Factor	Assessment
Financial Health	Very strong — $1B+ Series D at $26B; Lux, General Catalyst, 8VC, Founders Fund^[2]
Market Position	Leader — first mover, ~$492M run-rate, doubled ARR post-Windsurf^[1]
Innovation Pace	Rapid — Devin 3.0, Devin Desktop, SWE-1.6, desktop QA via computer use
Community/Ecosystem	Growing — active HN presence, polarized practitioner sentiment^[9]^[8] — proprietary platform, no open source
Long-term Outlook	Positive — but $26B valuation demands sustained hypergrowth

The Windsurf acquisition transformed Cognition from a single-agent vendor into a platform company with real distribution. The open question is whether revenue can grow into the valuation before model-native competitors commoditize the agent layer.

Bottom Line

Devin pioneered the autonomous AI software engineer category and has converted first-mover status into real scale: ~$492M run-rate revenue, a $26B valuation, and deployments at Citi, Mercedes-Benz, Goldman Sachs, and the U.S. military.^[1]^[2]

The old objections have partially aged out. Pricing dropped from $500/month to a $20 entry point, and the platform now offers OpenAI, Claude, and Gemini models alongside proprietary SWE-1.6 — Cognition is meaningfully multi-model, even if not BYOK. What remains are execution questions: opaque usage metering, weakness on ambiguous work, and a valuation that prices in flawless growth.

Recommended for: Enterprises with clear, repetitive workloads (migrations, security fixes, test generation) that want a unified IDE + autonomous-agent platform from one vendor.

Not recommended for: Teams needing true BYOK/self-hosted models, deep multi-agent orchestration across existing tools, or predictable per-seat-only pricing.

Outlook: Cognition has graduated from "first mover under siege" to category heavyweight — Windsurf gave it distribution, the Series D gave it runway, and multi-model support neutralized a key objection. The risks are now scale risks: integrating three product lines, defending a $26B valuation, and out-executing model-native tools (Claude Code, Codex) that keep getting cheaper.

Research by Ry Walker Research • methodology

Sources