Continue the conversation — chat opens pre-seeded with the current signal, caps, and movement.
Category-defining autonomous agent for async task offload. First to market with 'AI software engineer' positioning. Operates differently from IDE-based tools — delegates entire tasks rather than pair-programming. Now authors ~89% of Cognition's own PRs (up from 25%).
May 27, 2026: Series E closed — $1B+ at $25B pre / $26B post-money (2.5x in eight months), ~$492M ARR run-rate, enterprise usage +50% MoM for six straight months. January 2026: Infosys partnership represents 'largest global deployment of agentic software engineering to date' — 320,000 employees, 59 countries, Fortune 100 client base. The December 2025 Windsurf acquisition ($250M) thesis is now realized — Devin runs embedded inside the Windsurf IDE (not a separate tab), with SWE-1.5/1.6, Codemaps, and the Agent Command Center shipped through early 2026.
Devin shines when you can fully offload well-scoped work: migrations, legacy modernization, test generation, security remediation. Mercedes-Benz modernized a legacy system in eight days that had been projected at eight months; Itaú now auto-fixes 70% of its security vulnerabilities with Devin; Nubank achieved 8-12x engineering efficiency and 20x cost savings on a 6M-line ETL migration; Visma doubled developer productivity and halved project costs.
The fundamental trade-off: 67% PR merge rate in controlled environments vs ~15% success on arbitrary complex tasks (Answer.AI). Task selection is critical. Best fit: teams with large maintenance backlogs, legacy systems, or migration projects who can tolerate async review workflows.
Adoption & Proof Points
- Infosys (Jan 2026): 'Largest global deployment of agentic software engineering to date.' 320,000 employees, 59 countries. Three deployment models: internal productivity, hybrid delivery pods, MSP model. Financial Services leading first wave.
- Nubank: 6M+ line ETL migration, 8-12x engineering efficiency, 20x cost savings.
- Goldman Sachs: Piloting alongside 12,000 developers, 3-4x productivity uplift on specific workloads.
- Visma: Doubled developer productivity, halved project costs during application modernization.
- Litera: 40% test coverage increase, 93% faster regression cycles.
- Windsurf: Post-acquisition combined enterprise ARR up 30%. Less than 5% customer overlap.
- Cognition internal: 25% of internal PRs now produced by Devin, targeting 50%.
- Independent validation: Answer.AI found 15% success on arbitrary tasks (3/20). Cognition acknowledges 'senior at understanding, junior at execution.' 14x faster Java migrations vs human engineers (vendor-reported). 20x efficiency on security vulnerability remediation (vendor-reported).
Risks & Limitations
- Performance variability: 15-67% success depending on task scope and clarity. Independent testing consistently shows low complex-task success rates vs vendor-reported merge rates on controlled tasks.
- Windsurf integration shipped: Acquired Dec 2025 ($250M, $82M ARR). Unified Devin-in-IDE experience now delivered — Devin runs embedded in Windsurf (not a separate tab), with SWE-1.5/1.6, Codemaps, Devin Review/Quick Review, and Agent Command Center live; combined enterprise ARR up 30%+.
- Autonomy without guardrails: Documented rabbit-hole behavior — spends days pursuing impossible solutions rather than recognizing blockers. Requires clear acceptance criteria and ACU limits.
- Security surface: Shell/browser access enables exfiltration. April 2025 disclosure (120+ days for fixes). Unrestricted internet by default.
- Cost unpredictability: ACU consumption varies widely. ~15 min active work per ACU at $2-2.25/ACU. Vague prompts waste resources. No pre-task cost estimation.
- Cloud-only: No local execution. Source code must leave premises. Customer Dedicated SaaS mitigates but does not eliminate.
- Model choice but no BYOK: Adaptive routing with selectable models (SWE-1.6 default; opus/sonnet/codex/gemini via /model) but no customer-supplied API keys.
- Missing certifications: SOC 2 Type II + ISO 27001:2022 + CCPA present, but no FedRAMP or HIPAA — limits most-regulated-industry adoption.
- Oversight required: PR review essential. Confidence Scores help but code quality not self-verifiable.
Capabilities & Integration
Agentic depth: Full autonomous operation in sandboxed cloud environment with shell, code editor, Chromium browser. Multi-Devin orchestration enables parallel task execution (up to 10 workers). 'Devin Manages Devins' (Mar 19) enables parent-child session orchestration. Interactive Planning allows human-AI collaboration on task breakdown. Devin 2.2 (Feb 24): 3x faster startup, new UI, end-to-end desktop testing via computer use. Fast Mode (2x speed, Feb 13). Confidence Scores for success prediction.
Devin Review: Free code review product. Logical diff grouping, copy/move detection, severity-ranked bug flags. Auto-review triggers. Auto-fix button for bugs (Apr 3). CI status checks and logs (Apr 3). GitHub Enterprise Server support.
Context & Memory: DeepWiki (50K+ repos indexed), Wiki v2 with stronger reasoning (Apr 8). AskDevin Q&A and Plan modes. Knowledge base persists tips across sessions. Snapshots save machine state. Smarter codebase search with recency ordering.
Playbooks & Automation: Reusable prompts with steps, success criteria, guardrails. Scheduled Devins for recurring automation. Batch sessions across hundreds of repos. Custom Slash Commands.
Integrations: MCP Marketplace (Datadog, Sentry, Linear, Figma, Stripe, Amplitude). Native Linear integration. Slack, Teams, GitHub, GitLab, Azure DevOps. V3 API with RBAC. ACP methods expanded (Apr 8). Devin CLI enhancements.