AI Coding: what are the top startups?

Who still counts as an AI coding startup today?

An AI coding startup ranking now has to exclude Cursor, Base44 and Windsurf because all three have already left the independent field.

Cursor is the biggest change. SpaceX recently completed its $60 billion acquisition of the company, removing what would otherwise have been an obvious contender for the top spot. Cursor remains one of the most important AI coding products, but it now belongs in the same broad discussion as Claude Code, GitHub Copilot and Codex: strategically important products owned by much larger companies.

The same cleanup applies further down the market. Wix acquired Base44 after its breakout in prompt-based app creation. Windsurf followed a more unusual path: Google hired its CEO, co-founder and several key employees, then Cognition acquired the remaining company and product. Keeping either company in a current startup leaderboard would distort the answer.

We also have to draw the boundary around “AI coding” carefully. For this ranking, we include companies whose core business directly changes how software is created, modified, reviewed or operated. That covers autonomous engineering agents such as Cognition and Factory, app-building platforms such as Lovable and Replit, verification companies such as CodeRabbit and Qodo, orchestration products such as Warp and Augment, and code-model companies such as Poolside when software engineering remains central to what they build.

That leaves a much smaller independent field than many AI coding rankings suggest.

Company	Current status	How we treat it
Cursor	Acquired by SpaceX for $60B	Excluded from startup ranking
Windsurf	Core team moved to Google; remaining assets acquired by Cognition	Excluded
Base44	Acquired by Wix	Excluded
Cognition	Independent	Included
Lovable	Independent	Included
Replit	Independent	Included
Poolside	Independent after major Nvidia licensing deal	Included

Are independent AI coding startups still beating Claude Code and GitHub Copilot?

Independent AI coding startups have lost the broad professional-assistant lead to Claude Code, so the best startups today are winning more specialized parts of software development.

JetBrains' latest Developer Ecosystem Survey gives us one of the clearest pictures of actual professional usage. Among more than 15,000 professional developers surveyed worldwide, 90% said they were using AI coding agents at work at least weekly and 68% were using them daily.

Claude Code currently stands far ahead of every individual product in that survey. Around 39% of professional developers said they used Claude Code at work during the latest survey period, up from 18% in January. GitHub Copilot was at 21%, Codex at 16% and Cursor at 12%. OpenCode reached 7%.

The trajectory is even more telling than the ranking. Claude Code gained 21 percentage points in a matter of months. Codex moved from 3% to 16%. Cursor went from 18% to 12%, while GitHub Copilot continued slipping from its earlier leadership position.

This changes how we should judge startups. A company trying to become “the AI assistant every programmer uses” now has to fight Anthropic, OpenAI and Microsoft directly. That is a brutal position.

The stronger independent companies are moving into places where raw model adoption tells us less. Cognition is getting enterprises to delegate complete engineering tasks. Lovable, Replit and Emergent are turning people outside engineering into software creators. CodeRabbit and Qodo sit between generated code and production. Factory, Warp and Augment are trying to coordinate entire fleets of agents.

That is where the startup race currently lives.

Is Cognition the top AI coding startup right now?

Cognition is the strongest independent AI coding startup we can verify today.

The financial trajectory is unusually hard to match. Cognition disclosed a $73 million revenue run rate before combining Devin with Windsurf. By May 2026, run-rate revenue had reached $492 million, while enterprise usage had grown more than tenfold since the beginning of the year. In its latest funding discussions, Bloomberg reported that Cognition was approaching a $1 billion annualized revenue run rate.

If that latest figure holds, Cognition has moved from $73 million to close to $1 billion in roughly a year. Even allowing for the addition of Windsurf, that is an order-of-magnitude change in commercial scale.

The customer mix strengthens the case. Cognition has named Goldman Sachs, Citi, Mercedes-Benz, Santander, Dell and NASA among its customers. These are useful references because Devin is increasingly sold for work that large engineering organizations actually accumulate: migrations, dependency upgrades, legacy modernization, bug fixing and other tasks that can absorb substantial engineering time.

Cognition has also become broader than Devin alone. The Windsurf deal gave it an interactive coding environment alongside its autonomous agent. More recently, Cognition bought the team behind Poke, which suggests it is investing in how users interact with agents as those agents become more capable.

The company is still exposed to a serious threat from Claude Code and Codex. JetBrains' adoption data shows how quickly model companies can take distribution when their coding products improve. Yet Cognition has already crossed the point where we can describe Devin as an interesting technical experiment. Enterprises are paying hundreds of millions of dollars at an annualized pace for the product family.

Among independent companies, nobody else currently combines that level of revenue, enterprise usage and autonomous-engineering focus.

If you want more recent data on this point, please see our latest AI code assistant market report.

Is Lovable catching Cognition in AI coding?

Lovable is currently the closest startup to Cognition in commercial scale, although Lovable is winning with a much broader group of software creators.

The revenue curve has been extraordinary. Lovable reached $100 million in annualized subscription revenue eight months after launch, passed $200 million by the end of 2025, crossed $400 million in February 2026 and reached a $500 million annualized run rate by June.

The latest financing confirms that investors still see plenty of growth ahead. Lovable raised another $400 million at a $13.3 billion valuation, roughly twice the valuation from its previous round. The company also says it now hosts around 60 million projects receiving roughly 900 million visits per month.

What makes Lovable particularly interesting is the change in what customers are building. The first wave of vibe coding was full of websites, prototypes and weekend experiments. Lovable is increasingly pointing to internal business software, e-commerce systems, operational tools and applications used by real companies.

Enterprise revenue remains the weaker part of the story. Forbes reported earlier this year that only around $20 million of Lovable's then-$400 million ARR came from enterprise customers. That means the business was still overwhelmingly driven by individuals, entrepreneurs and smaller teams at that point.

Lovable has been working to change that mix. It has expanded its Google Cloud relationship, added enterprise distribution and security integrations, and says its product is now used somewhere inside a majority of Fortune 500 companies.

We therefore rank Lovable just behind Cognition. Cognition has better evidence of deep engineering adoption inside large organizations. Lovable has already built a larger mass-market phenomenon, and its ceiling could ultimately be higher because its addressable user is anyone who needs software rather than only people who work in engineering.

Is Replit still one of the top AI coding startups?

Replit still belongs in the top tier because its combination of 50 million-plus users, enterprise adoption and built-in deployment gives it a position that newer AI coding startups cannot easily recreate.

Replit's turnaround is one of the more important stories in this market. The company had existed for years before AI agents changed its trajectory. It then went from roughly $2.8 million in annualized revenue to $150 million in less than a year.

The company has stopped publishing a precise current ARR figure, which makes a direct comparison with Cognition and Lovable harder. Replit says it is on track for a $1 billion run-rate business by year-end, and TechCrunch recently described its current run rate as tracking toward that level. We treat that as directional evidence rather than counting $1 billion as revenue already achieved.

The usage side is clearer. Replit says more than 50 million people use the platform and people inside 85% of the Fortune 500 have built with it. Visa recently disclosed that more than 1,000 of its own employees use Replit for prototyping and development.

Replit also controls more of the workflow than most competitors. Someone can describe an application, have the agent build it, connect a database, run the software, deploy it and keep the project hosted without assembling a separate stack. Integrations with Databricks and Microsoft Fabric are pushing the same experience into governed enterprise data.

Retention may turn out to be especially important. CEO Amjad Masad has said some enterprise cohorts reach roughly 300% net revenue retention once customers start expanding their Replit usage.

That combination keeps Replit firmly in our top three. Lovable currently has the cleaner revenue disclosure and faster recent commercial curve. Replit has a ten-year product foundation, much broader infrastructure and a very large installed user base. It would be difficult to argue that either company has decisively left the other behind.

Is Poolside suddenly a top AI coding startup after the Nvidia deal?

Poolside has jumped into the top group strategically because Nvidia is paying $6 billion to license its model-building technology while separately investing $1 billion at a $12 billion pre-money valuation.

That is one of the biggest recent transactions anywhere in AI coding, and it changes how we should treat Poolside.

According to an investor letter reported by Newcomer, The Information, Bloomberg and The Wall Street Journal, Nvidia will receive a non-exclusive license to Poolside's Model Factory technology. Nvidia is also expected to offer jobs to 109 Poolside employees involved in the company's Laguna models. Poolside's three founders remain with the startup, which continues to exist independently.

The structure is unusual. Nvidia is effectively paying several billion dollars for access to the machinery Poolside developed to build and improve coding models, then investing another $1 billion into the company that remains.

That says a great deal about the value of Poolside's technology. A $6 billion licensing payment is larger than the total funding or disclosed revenue of almost every application-layer AI coding startup. Nvidia intends to use Poolside's technology and talent around its Nemotron open-model effort, giving us much stronger external validation than another venture round would provide.

Poolside still has a ranking problem: we know far less about widespread paid use of its coding products than we know about Cognition, Lovable or Replit. Its strongest asset may be the model-development system underneath the product rather than an end-user interface developers adopt every day.

We therefore put Poolside high in the ranking, but behind the three companies with the clearest evidence of scaled software usage. Right now, Poolside looks more like a strategically valuable AI coding model company than the default coding product for a large user population.

If you want more recent data on this point, please see our latest AI code assistant market report.

Is Emergent already a serious rival to Lovable and Replit?

Emergent has become a real AI software-creation contender, with roughly $120 million in annualized revenue and more than 200,000 paying customers.

The speed of the initial breakout was remarkable. Emergent passed $100 million in annualized revenue only eight months after launch, when it said it already had more than six million users and roughly 150,000 paying customers. Its latest disclosed figures put annualized revenue at $120 million and paying customers above 200,000.

The customer profile helps explain why Emergent can coexist with Lovable and Replit. Around 70% of Emergent users reportedly had no previous coding experience earlier this year. Small businesses accounted for a large share of usage, with trucking companies, construction firms, factories and property managers building software that previously lived in spreadsheets, email or messaging tools.

That is a useful niche. These customers often have real software problems but little reason to hire a traditional engineering team.

The later growth curve deserves more scrutiny. The public snapshots move from more than $100 million annualized revenue in February to $120 million by July, while management separately said revenue had risen around 70% over four months. Those figures may use different measurement periods, but they do not give us the same clean progression we get from Lovable.

The difference in scale remains large as well. Lovable's latest disclosed run rate is more than four times Emergent's, while Replit has more than 50 million users and much deeper infrastructure.

Emergent has nevertheless moved past the stage where it can be dismissed as another vibe-coding launch. Two hundred thousand paying customers make it a meaningful software company already. We currently place it immediately below the top group.

Is Factory the AI coding startup people are underestimating?

Factory is the strongest smaller challenger in enterprise AI coding, and its reported six-month revenue trajectory suggests it could move up the ranking quickly.

Factory says its Droids are used daily by hundreds of thousands of developers at companies including Nvidia, Adobe, EY, Palo Alto Networks and Adyen. TechCrunch has also reported Morgan Stanley among its customers.

The growth claim deserves attention. Factory says revenue doubled month over month for six consecutive months. Compounded literally, six monthly doublings mean the ending monthly revenue would be 64 times the starting level.

We should resist turning that into a revenue estimate because Factory has never disclosed the starting denominator. A company going from $100,000 to $6.4 million in monthly revenue tells a very different story from one going from $1 million to $64 million. Still, maintaining six consecutive doublings is strong evidence that something changed sharply in customer demand.

Investors have reacted accordingly. Factory's valuation moved from roughly $300 million in its previous round to $1.5 billion when it raised $150 million this year.

The product is also positioned sensibly for the current market. Factory can route work between different foundation models and focuses heavily on enterprise controls rather than asking customers to standardize on one underlying AI provider. As model leadership keeps shifting between Anthropic, OpenAI, Google and others, that flexibility becomes useful.

Factory lacks the absolute revenue disclosure needed to challenge Cognition today. Its customer list, daily usage and growth rate still make it the enterprise AI coding startup most likely to surprise on the upside.

Is CodeRabbit beating Qodo in AI code review?

CodeRabbit is currently ahead of Qodo commercially, while Qodo has enough enterprise adoption and benchmark performance to remain a serious challenger.

CodeRabbit has given us the clearer business trajectory. In September 2025, the company reported more than $15 million in ARR and over 8,000 business customers. In its latest financing, CodeRabbit raised $143 million at a $1.5 billion valuation after Axios reported that revenue had grown more than fivefold year over year.

The product is expanding with the business. CodeRabbit began with AI pull-request reviews and has moved into what it calls Agentic Change Management, covering planning, change impact, security and the growing problem of deciding which machine-generated changes should actually reach production. BMW has said more than 1,000 of its developers use the product.

Qodo has a strong enterprise roster of its own, including Walmart, Nvidia, Red Hat, Intuit, Texas Instruments, Ford and Monday.com. The company raised $70 million this year, taking total funding to $120 million.

Qodo also has credible technical evidence. TechCrunch reported that Qodo reached 64.3% on Martian's Code Review Bench, more than ten percentage points above the next competitor and around 25 points ahead of Claude Code Review at that time. Benchmarks can move quickly, but that gap is large enough to take seriously.

The wider market looks increasingly attractive. JetBrains now finds 68% of professional developers using coding agents daily, while CodeRabbit's own research found more issues in AI-generated pull requests than human-written ones. Even if we discount vendor-produced studies, teams are clearly generating code faster and placing more pressure on review.

That gives CodeRabbit and Qodo a natural tailwind. We rank CodeRabbit higher today because the commercial evidence is stronger.

Measure	CodeRabbit	Qodo
Latest funding	$143M Series C	$70M Series B
Latest valuation	$1.5B	Undisclosed
Growth evidence	Revenue up more than 5x YoY	Enterprise footprint expanding; revenue undisclosed
Enterprise examples	BMW, Chegg, Groupon, Mercury	Walmart, Nvidia, Red Hat, Intuit, Ford
Current position	AI review leader	Strongest specialist challenger

If you want more recent data on this point, please see our latest AI code assistant market report.

Is Warp becoming a bigger AI coding company than Augment Code?

Warp currently has the stronger commercial momentum, while Augment Code has built one of the more ambitious enterprise orchestration products in the market.

Warp started with a modern terminal and has steadily moved toward a much larger idea: an environment where developers can launch, coordinate and supervise coding agents locally and in the cloud.

The latest numbers make that transition worth taking seriously. Warp has nearly one million developers and says more than 56% of the Fortune 500 use the product somewhere inside their organizations. According to an OpenAI case study, Warp's ARR grew 35-fold over the previous year, while enterprise revenue increased more than 500% from the fourth quarter of 2025.

Warp's own engineering workflow gives us another useful indicator. Around 90% of its internal pull requests are now created with agents. Its Oz platform is built around coordinating those longer-running agents rather than simply giving one assistant access to a terminal.

Augment has reached a similar conclusion from a different starting point. The company is phasing out standalone completion features for most non-enterprise customers after seeing developers move toward agents. Its Cosmos platform now runs persistent software agents triggered by pull requests, alerts, schedules and other events.

Augment's published customer roster includes Adobe, MongoDB, Pure Storage, DXC, Snyk, MoneyGram and Webflow. Internally, the company says its code output has tripled since January after rebuilding more of its development workflow around agent teams.

The missing piece is commercial scale. Augment has disclosed substantial funding and customer names but little current revenue data. Warp gives us a much clearer growth picture.

For now, Warp ranks higher. Augment remains one of the more credible companies to watch as the market moves from one developer using one agent toward engineering teams managing many agents at once.

Why isn't Magic ranked with the biggest AI coding startups?

Magic remains an important code-model startup, but we still have far more evidence about its research capacity than about a scaled commercial coding business.

The resources behind Magic are substantial. The company says it has raised $515 million from investors including Sequoia, CapitalG, Jane Street, Eric Schmidt, Nat Friedman and Daniel Gross. It now has access to thousands of Nvidia GB200 GPUs and is working on frontier code models with ultra-long context and large amounts of inference-time compute.

Its ambition is equally large. Magic wants to automate software engineering deeply enough that AI can eventually contribute to AI research itself. Its work on context windows reaching tens of millions of tokens is aimed at letting models reason over extremely large codebases and technical histories.

What we cannot currently verify is a Cognition-style enterprise revenue base, a Lovable-style mass paid-user business, or a Replit-style installed platform.

Poolside used to have a similar problem in our ranking. The recent Nvidia transaction changed that because an external customer is now willing to pay $6 billion to license Poolside's technology. Magic has yet to produce a comparable commercial validation.

That leaves Magic in the high-potential group rather than our current top ten. A strong product launch or major licensing agreement could change that quickly.

If you want more recent data on this point, please see our latest AI code assistant market report.

Are Lovable, Replit and Emergent really competing with Cognition and Factory?

Lovable, Replit and Emergent are increasingly forming a separate software-creation market from the professional engineering agents sold by Cognition and Factory.

The difference starts with the user. Cognition usually enters an organization where software engineers, repositories and development processes already exist. Devin takes work from that engineering system and tries to complete more of it autonomously.

Factory follows a similar route. Its customers already have large engineering teams, and Droids operate across development, testing, review and deployment.

Lovable, Replit and Emergent often enter before an engineering team exists. A founder needs a customer portal. An operations employee wants an internal dashboard. A trucking company wants shipment-management software. A salesperson wants a custom workflow. AI generates the application and increasingly handles the infrastructure around it.

Emergent has said around 70% of its users had no previous coding experience. Replit talks explicitly about giving software creation to people across every profession. Lovable has gradually shifted its message from building websites toward running parts of a business through custom software.

That expands the addressable market enormously. Professional developers number in the tens of millions globally. Knowledge workers and small businesses that could benefit from custom software number in the hundreds of millions or billions.

The two markets will overlap more over time. Replit already serves professional developers alongside non-coders. Lovable is moving into enterprises. Cognition believes people who never coded before will increasingly direct engineering agents.

For ranking purposes, however, we should keep the distinction in mind. Cognition currently leads autonomous professional engineering. Lovable leads prompt-driven software creation at commercial scale. Replit has the strongest position across both worlds.

Is AI coding revenue as good as it looks?

AI coding revenue is growing at extraordinary speed, but annualized run rates still make several of these companies look more mature than their operating histories really are.

Cognition approaching a $1 billion annualized run rate, Lovable reaching $500 million, Emergent reaching $120 million and Replit tracking toward a much larger business are remarkable figures. We are seeing startups reach software-revenue levels in months that historically took successful SaaS companies many years.

Annualized revenue is still a snapshot. A company generating $40 million during a strong month can describe that pace as $480 million annually without having collected $480 million during the previous twelve months. That convention makes sense when revenue is changing this fast, but it also magnifies short-term acceleration.

Costs create the second complication. AI coding products can consume enormous amounts of inference, especially when agents keep running for minutes or hours, inspect large codebases, retry failed approaches and use frontier models.

Base44 gave us a rare public example after Wix acquired it. Wix disclosed that Base44 entered 2026 around zero non-GAAP gross margin. After shifting more workloads toward its own specialized model, Wix said it expected Base44's gross margin to move toward roughly 60% later in the year.

That change tells us why companies such as Lovable are training internal models, why Factory is investing in model routing, why CodeRabbit is optimizing cheaper specialized models for high-volume tasks and why Poolside built its own model infrastructure from the beginning.

The leaders have clearly proved that customers will pay. The next test is how much cash they can keep while usage keeps exploding.

Can AI coding startups survive Claude Code, Codex and GitHub Copilot?

Yes, the strongest AI coding startups can survive Claude Code, Codex and GitHub Copilot, but their advantage has to come from owning a workflow that foundation-model companies do not automatically inherit.

Claude Code shows how quickly a model company can change the market. JetBrains measured its workplace adoption rising from 18% in January to 39% in the latest survey period. Codex moved from 3% to 16% over roughly the same interval.

A startup whose main edge is slightly better code generation has very little protection against that speed.

The companies ranking highest here have already moved deeper into workflows. Cognition is selling measurable engineering output and recently offered enterprise customers a productivity guarantee worth up to $10 million in usage if Devin fails to generate enough estimated engineering value. Replit owns the environment where applications are generated, run and hosted. Lovable increasingly combines generation, deployment, integrations and business-software workflows.

CodeRabbit and Qodo benefit from being relatively model-neutral. They can review code written by Claude Code, Codex, Devin, Cursor or a human. Factory and Warp are heading in a similar direction by orchestrating whichever models perform best for a task.

Poolside represents the most extreme version of moving down the stack. Nvidia's willingness to pay $6 billion for access to its Model Factory suggests the machinery used to create and improve coding models can itself become a strategic asset.

The market is therefore getting harder for generic AI coding wrappers and more interesting for companies that control infrastructure, enterprise context, verification, deployment or agent coordination.

So which AI coding startups are actually on top today?

Cognition is our #1 AI coding startup today, followed by Lovable and Replit, with Poolside moving sharply upward after its enormous Nvidia deal.

Cognition has the best overall combination of commercial scale and deep engineering usage. Its latest reported trajectory points toward a $1 billion annualized revenue run rate, while enterprise adoption has expanded across companies such as Goldman Sachs, Citi, Mercedes-Benz and Santander. Devin also sits directly in the part of the market moving fastest: autonomous execution of real engineering work.

Lovable takes second place. A $500 million annualized revenue run rate, 60 million projects, 900 million monthly visits to those projects and a fresh $13.3 billion valuation make it difficult to rank lower. Its weaker enterprise mix keeps Cognition ahead for now.

Replit comes third. Current revenue disclosure is less precise, but 50 million-plus users, adoption somewhere inside 85% of the Fortune 500 and an integrated creation-to-deployment platform give Replit unusually broad strategic depth.

Poolside is the biggest change in our current ranking. Nvidia's $6 billion non-exclusive technology license plus a separate $1 billion investment at a $12 billion pre-money valuation provides massive external validation. We place it fourth because user adoption and recurring software revenue remain much less visible than at the top three.

Emergent is fifth with around $120 million in annualized revenue and more than 200,000 paying customers. Factory follows in sixth. Factory is smaller on disclosed financial scale but looks particularly strong in enterprise engineering, with hundreds of thousands of daily developers and six consecutive reported monthly revenue doublings.

CodeRabbit takes seventh and currently leads the independent verification layer. Warp ranks eighth because nearly one million developers, 35-fold ARR growth and its move into agent orchestration give it one of the strongest trajectories among the smaller players.

Qodo comes ninth, with a strong enterprise customer base and unusually good independent code-review benchmark results. Augment Code closes our top ten. Its commercial scale is harder to see, but Cosmos puts the company directly in the emerging market for persistent multi-agent software development.

The ranking also tells us something broader about where AI coding is going. Only a few years ago, the obvious startup opportunity was a better editor or autocomplete product. Today, the strongest independent companies are scattered across autonomous engineering, app creation, code models, verification and agent orchestration.

That fragmentation is likely to continue. Claude Code and Codex are taking more of the generic coding-assistant layer, while startups are building businesses around the parts of software creation where enterprise context, infrastructure, governance and complete workflow ownership still matter.

Rank	Startup	Strongest position today	Evidence that stands out
1	Cognition	Autonomous enterprise engineering	~$1B reported run-rate trajectory; major enterprise customers
2	Lovable	Prompt-to-software creation	$500M annualized run rate; 60M projects; $13.3B valuation
3	Replit	End-to-end AI software platform	50M+ users; presence inside 85% of Fortune 500; tracking toward much larger run rate
4	Poolside	Code models and model-building infrastructure	$6B Nvidia license plus $1B investment at $12B pre-money valuation
5	Emergent	Software creation for SMBs and non-coders	~$120M annualized revenue; 200,000+ paying customers
6	Factory	Enterprise coding agents	Hundreds of thousands of daily developers; six consecutive monthly revenue doublings
7	CodeRabbit	AI code review and change control	Revenue up >5x YoY; $1.5B valuation
8	Warp	Coding-agent orchestration	Nearly 1M developers; 35x ARR growth; >56% of Fortune 500
9	Qodo	AI code verification and governance	$120M raised; major enterprise customers; strong review benchmarks
10	Augment Code	Multi-agent enterprise development	Strong enterprise roster; Cosmos orchestration platform

If you want more recent data on this point, please see our latest AI code assistant market report.