Signals Inbox·July 20, 2026·AI Agents
Devin vs Factory Droids: who will win?
Devin is winning today on commercial scale, enterprise adoption, distribution and accepted production code, while Factory Droids remains dangerous where long-running work, private infrastructure and model neutrality matter most.
We track AI agents daily. Want the market signals in your inbox?
Send me the signals →Devin is winning the race against Factory Droids today. Cognition has the larger business, deeper enterprise deployments, stronger developer distribution and far more evidence that customers actually merge the code its agent produces.
The commercial gap matters more than the latest benchmark gap. Factory has a better history on public full-agent testing, but Cognition can point to hundreds of millions in company-wide run-rate revenue, organization-wide rollouts and a repeated stream of accepted pull requests.
Windsurf may be Cognition’s hardest advantage to copy. It puts the editor, cloud agent, proprietary models and production feedback inside one system, so every delegated task can strengthen distribution and improve the product at the same time.
Factory’s best route is not to become a smaller Cognition. Its opening is to own the model-neutral layer for persistent, multi-day software work across private machines, security tools and several AI models.
Neither company leads AI coding overall. GitHub Copilot, Claude Code and Cursor are larger, but inside autonomous enterprise engineering, Devin has the clearest lead and Factory still has a credible way back.
Interested in AI agents?We can send you all the signals
Send me the signals → Delivered straight to your inboxQ1Why do people keep comparing Devin and Factory Droids?
Devin and Factory Droids are compared because both want companies to hand complete software jobs to AI, rather than use AI only to suggest code inside an editor.
A developer can give either agent a ticket, repository or plain-English request. Devin or Droid can then explore the codebase, edit several files, run commands, test the result and return finished work for review. Both companies also sell enterprise controls for running many agents across private repositories and internal systems.
Their strategies now overlap across most of the software workflow. Cognition has expanded Devin through cloud agents, Devin Desktop, the Windsurf editor and its own coding models. Factory has widened Droid into code review, security, recurring automation and incident response.
The real split is what each company wants to control. Cognition is building an integrated AI engineering environment around Devin. Factory wants Droid to remain a flexible layer across different models, machines and developer tools.
So yes, they are direct rivals today. They may not be building toward the same destination.
Q2Why is the Devin vs Factory Droids race difficult to call?
Devin is winning commercially today, while Factory Droids still has a credible case on long-running work, model flexibility and independent technical testing.
The available numbers are uneven. Cognition publishes revenue, enterprise usage, merged pull requests and detailed customer outcomes. Factory discloses customer names, developer usage and rapid growth claims, but it still does not reveal current revenue.
The technical evidence measures different things. Factory has placed full Droid configurations on public leaderboards. Cognition reports strong results for its SWE models, although some tests use Cognition’s own harness or benchmarks and do not represent a direct Devin-versus-Droid contest.
Windsurf complicates the commercial comparison too. Cognition’s revenue now includes an established AI editor alongside Devin, while Factory remains closer to a pure agent platform. Total company revenue shows which business is larger, but not Devin’s exact share of the autonomous-agent market.
The answer comes down to four questions: who has more paying demand, whose code reaches production, which architecture handles difficult work better and which company is building the harder advantage to copy.
Q3Is Devin really much bigger than Factory Droids today?
Devin is operating at a much larger commercial scale than Factory Droids today, even after we separate Cognition’s acquired Windsurf business from Devin itself.
Cognition said in its latest funding announcement that the company had reached $492 million in annualized run-rate revenue, with enterprise usage increasing more than tenfold since the beginning of the year. That figure covers Cognition’s full product portfolio, so it should not be described as pure Devin revenue.
The last clean standalone number showed Devin at $73 million in ARR. Windsurf then added $82 million in ARR, more than 350 enterprise customers and hundreds of thousands of daily active users. Fewer than 5% of the two companies’ enterprise accounts overlapped before the acquisition, suggesting Cognition added a largely new customer base.
Factory has never disclosed an equivalent current number. The Financial Times previously reported that Factory was targeting $20 million to $25 million in annual contracts by the end of 2025. Factory later said revenue had doubled every month for six consecutive months, without revealing where that growth started or ended.
Even if Factory reached the top of its earlier target, Cognition’s current company run rate would be almost 20 times larger. The definitions differ, but this is not a close commercial race.
Commercial scale: Devin and Factory Droids
| Commercial measure | Devin and Cognition | Factory Droids | What it tells us |
|---|---|---|---|
| Latest disclosed run-rate revenue | $492 million company-wide | Not disclosed | Cognition is much larger and more transparent |
| Last clean standalone agent figure | Devin reached $73 million ARR | Factory targeted $20 million to $25 million in annual contracts | Devin was already several times larger before Windsurf |
| Acquired distribution | Windsurf brought $82 million ARR and 350+ enterprise customers | No comparable acquisition | Cognition accelerated through consolidation |
| Recent usage disclosure | Enterprise usage increased more than 10x | Hundreds of thousands of developers reportedly use Droids daily | Factory shows reach; Cognition shows stronger monetization |
We track AI agents daily. Want the market signals in your inbox?
Send me the signals →Q4Is Factory Droids growing faster than Devin right now?
Factory Droids may be growing faster in percentage terms, but Devin appears to be adding far more revenue and enterprise usage in absolute terms.
Factory’s six reported monthly doublings imply a 64-fold increase across the full period. That is an exceptional curve. The missing starting number still matters: a 64-fold increase can produce a modest business or a very large one depending on the base.
Factory’s recent hiring suggests it now wants to turn early demand into a repeatable enterprise business. The company appointed Marcello Gallo as chief revenue officer after senior sales roles at Sigma, Moveworks and MongoDB. That usually happens when founder-led selling and inbound demand are no longer enough.
Cognition’s growth can be measured more clearly. Devin previously rose from $1 million to $73 million in ARR within nine months. More recently, Cognition reported tenfold enterprise usage growth across a far larger customer base.
Factory has the more explosive recent percentage claim. Devin has the stronger growth story overall because each new customer and deployment is being added to a much larger base.
Q5Does Devin have better enterprise customers than Factory Droids?
Devin currently has the stronger enterprise customer record because Cognition shows more company-wide deployments and more evidence of measurable operational results.
Factory already works with serious organizations. It names Nvidia, Adobe, EY, Palo Alto Networks, Adyen, Blackstone and Wipro among companies using its software-factory products. Factory also says hundreds of thousands of developers now use Droids daily.
The public case studies remain narrower. Nav reported a 60% reduction in context-switching time and faster feature cycles. Empower said Droid reduced incident-response time by up to 40%. You.com introduced Droid into its regular code-review process.
Cognition shows deployments spreading across entire engineering organizations. Mercedes-Benz said Devin reduced one legacy-modernization project from eight months to eight days. Itaú reports that Devin is used by 75% of its technology teams, has documented more than 300,000 repositories and automatically fixes 70% of selected security vulnerabilities.
Nubank used Devin on a migration involving a multi-million-line ETL system that would otherwise have required coordination across more than 1,000 engineers. Cognition also names Citi, Goldman Sachs, Santander, Dell, the U.S. Army and the U.S. Navy as customers.
These are vendor-selected examples, not average results. Still, the pattern is fairly clear: Cognition has more regulated customers, more organization-wide usage and more projects where Devin has entered core engineering operations.
Q6Are companies really shipping code written by Devin and Factory Droids?
Companies are clearly shipping Devin-written code today, while Factory has not published enough product-wide data to show how much Droid work reaches production.
Cognition says 89% of the code committed by its own engineers now comes from Devin, with the remainder produced through local agents inside Windsurf. During one recent week, Cognition merged 659 Devin pull requests into its own codebase, compared with 154 during its best week in 2025.
Gumroad offers a useful external example. Devin had produced 1,583 merged pull requests there, averaged around ten merged PRs per day and achieved an 85% merge rate. It became the top contributor in five of Gumroad’s seven most active repositories.
Factory has shown that Droid is used in serious production environments. Groq runs Droid through its own inference infrastructure. Chainguard engineers have kept individual Droid sessions alive for weeks. You.com uses it during code review.
Factory has not disclosed an overall merge rate, total merged pull requests or another comparable measure of accepted output. Daily usage shows interest. It does not show how often Droid’s work survives review and enters production.
Devin wins this one comfortably because Cognition can point to a repeated flow of code that companies actually merge.
Interested in AI agents?We can send you all the signals
Send me the signals → Delivered straight to your inboxMeta is preparing Hatch, a consumer AI agent, within weeks
Hark is building Handoff to outperform humans across the internet
Ex-Cursor engineer claims GrokBots handle 90% of his routine
AI agents now burn nearly 5x more tokens than humans
27B Faraday beats OpenAI and Anthropic at paper replication
Binance’s Agent OS now connects ChatGPT, Claude Code, and Cursor
OpenAI’s agent users hit 20 million, enterprise revenue jumps 50%
Two-month-old River AI just raises $1.1B for personal agents
Meta launches a 30B open-weight model for local AI agents
Cursor can now read, write, and act across your Google Workspace
Paul Graham: AI might never replace startup founders
Q7Is Factory Droids technically better than Devin today?
Factory Droids has the stronger history on independently listed agent benchmarks, but recent results no longer show a clear technical lead over Devin.
Factory built much of Droid’s early reputation through Terminal-Bench, which tests agents on practical terminal tasks. Droid set a record at 58.75% in 2025. A later configuration using GPT-5.3-Codex reached 77.3% on Terminal-Bench 2.0.
That score has since fallen to tenth place as newer systems moved into the low-to-mid 80s. Factory’s old leaderboard lead says much less now than it did when the result was published.
Cognition has moved faster at the model layer lately. Its SWE-1.7 model scored 81.5% on Terminal-Bench 2.1, 77.8% on SWE-Bench Multilingual and 42.3% on FrontierCode 1.1. Cognition designed the model for asynchronous software tasks and added self-compaction so it can continue working after its initial context fills up.
These are not clean head-to-head results. Cognition ran some evaluations through its own framework, and FrontierCode is a Cognition-created benchmark. Factory’s full-agent results have generally been easier to inspect on independent leaderboards.
Factory keeps a narrow edge for publicly verified agent testing. Cognition looks stronger in proprietary model progress and real production output.
Technical evidence: Factory Droids vs Devin
| Technical question | Factory Droids | Devin and Cognition | Current judgment |
|---|---|---|---|
| Independent agent testing | Droid scored 77.3% on Terminal-Bench 2.0 | No comparable current full-Devin listing | Factory |
| Current leaderboard strength | Droid’s result now ranks tenth | No full-product entry | Factory’s previous lead has narrowed |
| Proprietary coding models | Mostly relies on external and open models | Cognition trains SWE models, including SWE-1.7 | Cognition |
| Production acceptance | Strong customer examples, no overall merge rate | Large disclosed volumes of merged code | Devin |
| Overall technical position | Better public harness record | Better current model and production story | Too close for a clean winner |
Q8Can Factory Droids handle longer jobs than Devin?
Factory Droids currently has the clearer system for software projects that run for hours or days without constant human supervision.
Factory Missions breaks a large request into milestones, assigns separate workers, validates completed work and creates follow-up tasks when checks fail. Factory has published examples including a 33.8-hour COBOL-to-Java migration, a 24.2-hour memory-leak investigation and a desktop application built over roughly 30 hours.
Droid Computers gives those agents a persistent environment. A Droid can return to the same filesystem, packages, credentials, local services and process state rather than reconstructing its setup for every new session.
Devin also supports parallel and asynchronous execution. Teams can launch sessions from Slack, Jira, Linear, APIs and scheduled automations, then supervise several Devins at once. Cognition uses this heavily inside its own engineering organization.
SWE-1.7 improves Devin’s long-horizon capabilities too. Cognition trained it with self-compaction and ran training tasks lasting up to six hours. That helps Devin preserve useful context during longer jobs, although it still differs from Factory’s explicit multi-day orchestration.
Factory has simply demonstrated this capability more clearly. Missions and persistent computers are built for the point where agents often break down: several workers, changing environments and projects that outlive one context window.
Q9Did Windsurf give Devin the better developer workflow?
Windsurf currently gives Devin the stronger daily developer workflow because Cognition controls both the editor where engineers work and the cloud agent that continues after they leave.
The acquisition brought Cognition $82 million in ARR, more than 350 enterprise customers and hundreds of thousands of daily active users. The low overlap between Devin and Windsurf customers gave Cognition a new distribution channel rather than another route into the same accounts.
The product fit was obvious. Devin began as a cloud worker that engineers could send away to complete tasks. Windsurf gave Cognition the interactive environment where those same engineers read, write and debug code themselves.
Devin Desktop now connects the two modes. A developer can edit locally, launch a cloud agent, inspect its work and take over without switching to an unrelated product. Cognition can also sell codebase search, local assistance, autonomous execution and review through one environment.
Factory supports a CLI, desktop application, browser, SDK and private machines. Its latest desktop releases can preview code, websites, PDFs, presentations and spreadsheets beside the Droid session. Factory offers more freedom over where work runs.
Cognition still owns the stronger daily workflow. Windsurf puts Devin inside an interface developers already use for hours, while Factory has to create that habit itself.
We track AI agents daily. Want the market signals in your inbox?
Send me the signals →Q10Is Devin or Factory Droids safer for banks and government teams?
Devin currently has the stronger record with regulated institutions, while Factory Droids offers more flexibility for companies that need full control over their infrastructure.
Cognition’s platform is listed as FedRAMP High In-Process. Devin Desktop and Devin CLI already carry federal and defense approvals including FedRAMP High and several Department of Defense impact levels. Cognition also names deployments across the U.S. Army, U.S. Navy, Treasury Department and NASA’s Jet Propulsion Laboratory.
Factory supports zero-data-retention configurations, dedicated inference, customer-managed encryption keys, data residency and on-premise deployment. A company can also restrict which models receive particular data and run Droids on privately managed machines.
Factory has added more security directly into the product. Droid Shield 2.0 is designed to detect exposed credentials, while automated security review checks agent-written code before it is merged. Droid Shield remains a research preview, so it carries less weight than a government authorization or an established regulated deployment.
Devin leads on proof. Factory may still be the better fit when air-gapped infrastructure, model choice or customer-controlled compute matters more than existing certifications.
Q11Are Devin and Factory Droids still trying to build the same company?
Devin and Factory Droids still compete for autonomous coding work, but their recent releases show that Cognition and Factory are increasingly building different companies.
Cognition is moving deeper into an integrated engineering stack. SWE-1.7 gives it a proprietary model. Devin Fusion routes different parts of a job between models. Devin Desktop connects cloud agents with the Windsurf editor. Its government approvals open additional regulated markets.
Factory is moving across more parts of software operations. Its recent products cover incident response, security review, persistent computers and recurring automations. A Droid can react to an alert, investigate the issue through observability tools, prepare a fix and save what it learned for the next incident.
Cognition wants to control the model, agent, editor and enterprise deployment around Devin. Factory wants Droid to work across coding, review, security, monitoring and other workflows while remaining compatible with several models and infrastructure setups.
That gives Factory a better chance than a direct feature race would. Cognition has more revenue, capital and distribution. Factory can compete by becoming the model-independent operating layer used across several agents, including products that rival Droid.
Q12Does Cognition have a better model strategy than Factory?
Cognition currently has the stronger model strategy because Devin can use outside models, route between them and rely on models trained by Cognition itself.
Factory’s neutrality still has real value. Droid can use leading commercial models and open-weight alternatives. Factory Router selects models according to quality, speed, cost and availability, while enterprise customers can decide which models may access sensitive data.
Factory benefits whenever OpenAI, Anthropic, Google or an open-source laboratory releases a stronger model. It does not need to win the expensive race to train the best foundation model.
Cognition now combines that flexibility with its own model program. SWE-1.7 runs across Devin Web, Desktop and CLI. Devin Fusion can switch models during one session rather than committing the entire job to a single model from the beginning. Cognition said Fusion maintained similar benchmark performance while reducing model costs by 35% in its tests.
Factory carries less model-development risk. Cognition has more ways to improve performance, reduce cost and keep control when an outside provider changes its pricing or product access.
Interested in AI agents?We can send you all the signals
Send me the signals → Delivered straight to your inboxQ13Which advantage is harder to copy: Devin’s scale or Factory’s neutrality?
Devin’s scale is currently harder to copy than Factory Droids’ neutrality because Cognition’s customer base, accepted production code, editor distribution and proprietary models strengthen one another.
Cognition sees where Devin succeeds, where engineers intervene, which pull requests get accepted and which jobs repeatedly fail. Those patterns can improve task selection, model routing, onboarding and future model training.
Windsurf adds another layer. Cognition can observe when developers work directly inside the editor, when they delegate a task and what they change before accepting the result. Partnerships with Infosys and Cognizant then bring Devin into large transformation projects that Cognition could not reach as quickly through direct sales alone.
Cognition also has much stronger evidence that Devin-generated work reaches production. Accepted output gives the company more useful feedback than raw prompts, generated code or daily-user counts.
Factory’s neutrality remains attractive. Companies can choose different models, machines and autonomy levels without placing the entire workflow inside Cognition’s environment. That could become valuable as enterprises standardize on several agents rather than one.
For now, neutrality is easier to reproduce than Cognition’s feedback loop. A competing platform can add model choices and private deployment options. Recreating Cognition’s distribution, enterprise relationships and volume of accepted code would take much longer.
Q14Can Factory afford to fight Cognition for several more years?
Factory has enough capital to remain a serious competitor, but Cognition currently has far more room to train models, subsidize usage, acquire products and survive mistakes.
Factory raised $150 million at a $1.5 billion valuation after an earlier $50 million round at a $300 million valuation. Its value increased fivefold between the two financings, and its investors include Khosla Ventures, Sequoia, Nvidia, JPMorgan, Blackstone and Insight Partners.
Cognition raised more than $1 billion at a $26 billion valuation. Its latest round alone was over six times larger than Factory’s, while the valuation was roughly 17 times higher.
The funding gap matters because autonomous software agents consume expensive inference before anyone knows whether the final output will be accepted. Long jobs may require several models, large codebase contexts, repeated test runs and a few dead ends.
Factory may run a leaner operation by relying more heavily on outside models. It has enough backing to keep investing in Droid and pursuing large customers.
Cognition can fund more strategies at once: proprietary model training, enterprise sales, lower pricing, acquisitions, government compliance and international expansion. That gives Devin more room for error if the market becomes expensive or crowded.
Q15Are Devin or Factory Droids leading the wider AI coding market?
Neither Devin nor Factory Droids currently leads the overall AI coding market, although Cognition operates much closer to the largest players.
Microsoft’s latest disclosures put GitHub Copilot above 4.7 million paid subscribers and nearly 140,000 customer organizations. Its position inside GitHub gives Copilot a distribution advantage that neither Cognition nor Factory can easily match.
Anthropic said Claude Code had passed $2.5 billion in run-rate revenue, with weekly active users doubling since the beginning of the year. Cursor had already crossed $1 billion in annualized revenue and said more than 70% of the Fortune 500 used its products.
These companies started from different positions. Copilot dominates through GitHub. Cursor and Claude Code built strong interactive developer workflows. Devin and Factory focus more heavily on asynchronous agents that can receive an entire unit of work.
Inside that narrower market for autonomous enterprise engineering, Devin is ahead of Factory Droids. Across AI coding as a whole, both are still challengers.
We track AI agents daily. Want the market signals in your inbox?
Send me the signals →Q16Who is winning between Devin and Factory Droids right now?
Devin is clearly winning today because Cognition has greater commercial scale, deeper enterprise deployments, stronger distribution and much more proof that customers are merging the work its agent produces.
Factory leads in a few areas that could matter later. Droid has the better history on independent agent benchmarks. Factory has also made a stronger public case for persistent machines, multi-day orchestration, private infrastructure and model-neutral deployment.
Those advantages have not yet translated into comparable commercial evidence. Factory still does not disclose current revenue or an overall measure of how much Droid-generated work reaches production.
The three factors carrying the most weight are commercial scale, accepted production output and developer distribution. Benchmarks can change quickly when a new model appears. Customer workflows, enterprise contracts and a large installed base are much harder to overturn.
Factory could change the answer by showing substantial paid revenue, repeated organization-wide deployments and clear demand for an independent control layer above several coding agents. It also needs an aggregate measure connecting Droid usage with accepted production work.
Cognition has work to do too. Devin must remain economically attractive as sessions become longer and more computationally expensive. Cognition also needs to integrate Windsurf cleanly and prove that usage extends beyond a small group of unusually large customers.
Our verdict is direct: Devin is winning the market for autonomous enterprise software engineering today. Factory Droids remains a credible challenger because its open architecture and long-running agents attack genuine weaknesses in the category. Factory’s best chance comes from defining a broader, model-neutral software-factory market before Cognition closes that opening.
Devin vs Factory Droids: final scorecard
| Criterion | Who is ahead today? | How clear is the gap? | Why it carries weight |
|---|---|---|---|
| Commercial scale | Devin | Very clear | Cognition operates at a much larger disclosed revenue base |
| Recent percentage growth | Factory Droids | Unclear | Factory reports six consecutive monthly doublings but omits the starting value |
| Enterprise deployment depth | Devin | Clear | Cognition shows more organization-wide and regulated deployments |
| Accepted production code | Devin | Clear | Cognition publishes large merged-PR volumes and strong customer examples |
| Independent benchmark history | Factory Droids | Narrow | Droid has the better public full-agent testing record |
| Long-running orchestration | Factory Droids | Moderate | Missions and persistent computers are built around multi-hour and multi-day work |
| Developer distribution | Devin | Clear | Windsurf gives Cognition an established editor and daily user base |
| Regulated adoption | Devin | Clear | Cognition has stronger approvals and named government deployments |
| Infrastructure flexibility | Factory Droids | Moderate | Factory offers broader on-premise and model-control options |
| Model strategy | Devin | Moderate | Cognition combines outside models, dynamic routing and proprietary models |
| Financial capacity | Devin | Very clear | Cognition has raised far more capital at a much higher valuation |
| Wider AI coding leadership | Neither | Clear | GitHub Copilot, Claude Code and Cursor remain larger overall |
| Final verdict | Devin | Clear, but reversible | Factory must turn its architectural strengths into disclosed commercial scale |
We track AI agents daily. Want the market signals in your inbox?
Send me the signals →We broke the comparison into the dimensions that most clearly show competitive strength: commercial scale, growth, enterprise deployment, accepted production work, technical performance, long-running execution, developer distribution, security, infrastructure flexibility, model strategy and financial capacity.
We used the freshest available company disclosures, customer outcomes, independent benchmark results, product releases, official government records and tier-1 reporting. We calculated ratios, implied growth and relative scale when the isolated figures did not make the gap clear.
For commercial scale, we treated Cognition’s $492 million figure as company-wide revenue rather than pure Devin revenue. We used Devin’s last clean standalone $73 million ARR figure and Factory’s earlier $20 million to $25 million annual-contract target as the clearest direct historical anchors.
For technical performance, we used Terminal-Bench to judge Factory’s public full-agent testing and Cognition’s SWE-model results to judge model progress. We did not combine them into a fake head-to-head score. Production acceptance was assessed separately through merged pull requests, merge rates and customer deployment evidence.
Customer case studies show what the products have demonstrated, not what every customer should expect. In the final verdict, we gave the most weight to paid demand, accepted production code, organization-wide adoption, durable distribution and advantages that would take years rather than months to reproduce.
Key sources used for this analysis include: Cognition’s latest funding and revenue announcement, Cognition’s standalone Devin growth disclosure, Cognition’s Windsurf acquisition announcement, Cognition’s internal Devin usage report, Devin’s annual performance review, Cognition’s SWE-1.7 release, the official FedRAMP Marketplace, Factory’s Series C announcement, the independent Terminal-Bench 2.0 leaderboard, Factory Missions, Droid Computers, Factory’s enterprise infrastructure documentation, Microsoft’s GitHub Copilot disclosure, Anthropic’s Claude Code disclosure, and Cursor’s company growth announcement.
Building or investing in AI agents?We can send you all the signals
Send me the signals → Delivered straight to your inbox