Which AI agent startup is growing the fastest?

Why is the AI agent startup race so hard to rank right now?

The AI agent startup race is unusually hard to rank today because several companies are growing at absurd speeds, but they are doing it from very different starting points.

Cognition is already generating more than $900 million in annualized revenue. Manus has climbed into the $400 million to $500 million range. Harvey has reached roughly $350 million ARR. Genspark reported around $250 million. Sierra is around $200 million. Legora has reached roughly $150 million, while Decagon is around $100 million.

Then there is Wonderful. Its revenue run rate is only about $70 million, but that figure was roughly $1 million a year earlier.

Those numbers create two very different races. Wonderful looks almost unbeatable if we rank companies by percentage growth. Cognition looks much stronger if we care about how many new revenue dollars a company can add after it has already become large.

There is another complication. AI coding tools, legal platforms and customer-service systems are all becoming more agentic. Cursor would have been impossible to ignore after crossing $2 billion in annualized revenue, for example, but SpaceX completed its acquisition of the company in August. Cursor therefore no longer belongs in a ranking of independent startups.

So we need to be precise about both words in the title: what counts as an AI agent startup, and what kind of growth deserves to be called the fastest.

What should actually count as an AI agent startup?

For this comparison, an AI agent startup has to sell software that can carry out multi-step work on the customer's behalf rather than mainly generating an answer.

Cognition qualifies because Devin can take a software task, work through files and tools, write and test code, and return completed work. Manus and Genspark qualify through broader computer and knowledge-work agents. Sierra, Decagon and Wonderful run customer-facing workflows. Harvey and Legora increasingly let legal teams build agents that review documents, research issues and produce structured work across several steps.

We would leave out companies such as Anthropic even though Claude powers millions of agentic workflows. Anthropic's main business is selling frontier models, so putting it beside Sierra or Cognition would turn the article into a comparison of very different businesses.

The same filter removes traditional SaaS companies that happen to have added an agent button. Agentic work needs to be central to what customers are buying.

This definition is broad enough to capture where the market is actually going without stretching “AI agent startup” until almost every AI company qualifies.

How should we decide which AI agent startup is growing the fastest?

The best measure of AI agent growth combines percentage growth, absolute revenue added and the size of the business before the growth happened.

Percentage growth on its own heavily favors tiny companies. Going from $1 million to $10 million means 900% growth, even though the company added only $9 million.

Absolute growth creates the opposite problem. A company that is already huge can add more dollars while growing much more slowly.

For this article, we care most about companies that are still multiplying quickly after reaching meaningful commercial scale. A move from $100 million to $250 million tells us more than a move from $2 million to $10 million. A move from roughly $500 million to $900 million is harder again because each additional percentage point represents several million dollars of annualized revenue.

Time also matters. We should treat an explosive three-month period differently from a growth rate sustained for a full year.

That leaves us with a fairly demanding test: how much did the company grow, how quickly did it happen, and how large was the business when it happened?

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

Which AI agent startups are growing fast enough to have a real claim?

Cognition currently has the strongest combination of scale and speed, while Genspark, Manus and Wonderful each win a narrower version of the growth race.

We compared the latest usable revenue figure with the closest earlier figure that appears reasonably comparable. The final column is particularly useful because it converts each company's increase into the approximate amount of annualized revenue added for every month that passed.

That calculation exposes a huge gap. Cognition has recently been increasing its annualized revenue base by roughly $136 million per month. Genspark's exceptional early-year burst worked out to about $50 million per month. Manus has been adding roughly $38 million to $50 million per month depending on whether we use the bottom or top of its latest reported range.

Wonderful's seventyfold increase looks spectacular in percentage terms, yet it translates into roughly $6 million of additional annualized revenue per month because the company started from such a small base.

AI agent startup	Revenue trajectory	Approx. period	Annualized revenue added per month
Cognition	~$492M → $900M+	~3 months	~$136M
Genspark	~$100M → $250M	3 months	~$50M
Manus	~$100M → $400–500M	~8 months	~$38–50M
Harvey	~$195M → $350M	~7 months	~$22M
Legora	~$50M → $150M	~6 months	~$17M
Sierra	~$130M → $200M	~5 months	~$14M
Decagon	~$44M → $100M	~7 months	~$8M
Wonderful	~$1M → $70M	~12 months	~$6M

Is Cognition the fastest-growing AI agent startup today?

Yes. Cognition currently has the best claim because its revenue almost doubled in roughly one quarter even though the company was already approaching half a billion dollars in annualized revenue.

Cognition disclosed $492 million of run-rate revenue in late spring. More recent reporting from The Information put the company at around $900 million, equivalent to roughly $75 million of revenue in its latest month multiplied by twelve. Bloomberg subsequently reported that annualized revenue had moved above that level.

The jump works out to roughly 83% in around three months. Compound that over the period and Cognition was growing at roughly 22% per month.

The dollar increase is even more striking. Cognition added approximately $408 million of annualized revenue in that short window. Sierra's entire current ARR is about half that amount. Decagon's is roughly one-quarter.

Cognition has also shown that Devin usage itself is rising quickly. When the company reported the earlier $492 million figure, it said enterprise usage of Devin had grown more than tenfold since the start of the year. CEO Scott Wu told TechCrunch that enterprise usage had been increasing around 50% month over month during the previous six months.

That is why Cognition sits at the top of our ranking. The company has kept something close to early-startup growth rates after reaching a revenue level where growth normally becomes much harder.

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

Does Windsurf make Cognition's growth look bigger than it really is?

Windsurf definitely inflated Cognition's starting scale, but the acquisition cannot explain the company's recent surge.

Cognition acquired the remaining Windsurf business in 2025. Around that time, Windsurf was generating roughly $82 million ARR, while Devin was around $73 million.

Anyone claiming that Devin organically went from $73 million to more than $900 million would therefore be overstating what happened. The business changed materially when Windsurf joined it.

But look at what happened after the companies had already been combined. A rough $155 million mid-2025 revenue base became $492 million by late spring this year, followed by another sharp acceleration during the next quarter.

Cognition has also been increasingly folding Windsurf and Devin into one coding platform. Customers can move between interactive coding and more autonomous work instead of buying two completely separate products.

So the Windsurf acquisition deserves a real adjustment when we describe Cognition's long-term growth. It does much less to weaken the recent growth claim, because the biggest jump happened well after Windsurf was already inside Cognition.

Is Manus growing faster than Cognition?

Manus is still one of the fastest-growing AI agent companies in the world, but the freshest numbers no longer show it clearly growing faster than Cognition.

Manus was generating around $100 million in annualized revenue when Meta acquired the company in December. The Information now puts its annualized revenue run rate between $400 million and $500 million.

That means Manus multiplied its revenue roughly four to five times in about eight months. Using the two ends of that range gives us compound monthly growth of roughly 19% to 22%.

Cognition's latest comparable period comes out at roughly 22% a month. The two are therefore very close at the top end, while Cognition has added far more dollars because it started from a much larger base.

Manus also carries one unusual complication. Its fastest expansion happened while Meta owned the company. Chinese regulators later forced the transaction to be unwound, and Manus has now returned to independent operation, with earlier investors involved in buying it back.

We cannot tell how much Meta contributed to the acceleration through infrastructure, credibility, distribution or simple attention. The Information has also noted that the precise reasons for the increase are unclear.

Still, Manus had already reached $100 million before most of that period began, so Meta cannot take credit for creating the business from nothing. The company sells subscriptions ranging from roughly $20 to $200 a month and has continued expanding its agent product while going through an ownership situation that would have distracted almost any startup.

Manus belongs near the top of the ranking. The fresh data just makes Cognition's claim stronger than it looked a few months ago.

Did Genspark have the fastest AI agent growth burst?

Genspark produced the fastest short revenue burst we found among agent startups that had already crossed $100 million ARR.

Genspark started as an AI search product before shifting toward Super Agent, which can research, make presentations, analyze data, create software and perform other multi-step tasks.

The initial takeoff was extraordinary. OpenAI reported that Super Agent reached $36 million ARR only 45 days after launch with a team of roughly 20 people and no paid advertising.

Genspark later reached about $100 million ARR and then added another $150 million during the first three months of this year. Reuters reported the company's own figure of roughly $250 million ARR when it raised additional funding, alongside more than 6,000 companies that had signed up for Genspark for Business within six months.

Going from $100 million to $250 million in one quarter corresponds to roughly 36% compound monthly growth. That is faster than Cognition's latest three-month percentage rate.

The problem is freshness. We have not seen another equally credible revenue milestone since the $250 million disclosure. We know the company has continued shipping products and selling to businesses, but assuming the previous 36% monthly rate continued would quickly produce fantasy numbers.

Genspark therefore gets credit for the fastest verified three-month burst in our comparison. We would need a newer revenue figure before calling it the fastest-growing company today.

Is Wonderful actually the fastest-growing AI agent startup?

Wonderful is currently the percentage-growth champion, with its revenue run rate rising from roughly $1 million to about $70 million in a year.

A seventyfold increase is hard to put in normal SaaS terms. It works out to compound monthly growth of roughly 42%.

Wonderful has also moved well beyond the stage where a handful of customers can create a misleading percentage. The company says it now operates across more than 35 markets, employs around 650 people and helps enterprises automate workflows through agents and AI applications. Deutsche Telekom is among the customers publicly associated with the platform.

Its latest financing adds another useful piece of evidence. Wonderful raised $550 million at a $5 billion valuation, more than doubling its previous valuation in less than six months. The Wall Street Journal reported the roughly $70 million revenue run rate alongside that round.

Wonderful's forward-deployed strategy helps explain some of the speed. Instead of waiting for enterprises to implement the product themselves, engineers work closely with customers to connect agents to existing systems and workflows.

The catch is simply scale. Wonderful has added about $69 million of annualized revenue over a year. Cognition recently added several times that amount in one quarter.

If “fastest” means highest percentage growth, Wonderful wins. Once we require large-scale growth as well, Cognition moves back ahead.

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

Is Sierra or Decagon growing faster in customer-service AI agents?

Sierra is currently winning the customer-service AI agent race on scale, while Decagon is growing at a similar percentage rate and Wonderful is moving fastest from a smaller base.

Sierra went from roughly $130 million ARR at the end of last year to around $200 million by May, according to Sacra's estimates. The company had crossed $100 million only a few months earlier.

That progression is backed by unusually broad enterprise adoption. Sierra says more than 40% of the Fortune 50 use its platform and its agents have already handled billions of customer interactions. More recently, the company has expanded into long-running agents that can follow customers across several interactions rather than handling only one support conversation.

Decagon reached roughly $100 million in annualized revenue in July, according to Sacra, up from around $44 million at the end of last year. That is approximately 127% growth over seven months, equivalent to roughly 12% compounded monthly.

Sierra's recent growth works out in roughly the same range. The main difference is that Sierra got to $100 million earlier and has already built the next $100 million on top.

Wonderful remains the unusual third competitor. Its percentage growth dwarfs both companies, although its revenue base is still smaller and its implementation-heavy model looks different from Sierra's.

Customer-service AI agent startup	Latest revenue level	Recent growth	What separates it
Sierra	~$200M ARR	~$130M → $200M	Largest scale and deep Fortune 50 penetration
Decagon	~$100M annualized revenue	~$44M → $100M	Roughly matches Sierra's recent percentage pace
Wonderful	~$70M run rate	~$1M → $70M	Fastest percentage growth by far

Is Harvey or Legora growing faster in legal AI?

Legora is growing faster proportionally, but Harvey is still adding more revenue and remains the bigger legal AI company.

Harvey has reached roughly $350 million ARR, according to a very recent interview with co-founder Winston Weinberg in The Times. Sacra estimates that Harvey finished last year around $195 million.

That implies an increase of roughly $155 million in about seven months.

Legora's growth is faster from a smaller starting point. Sacra estimates it moved from around $50 million ARR at the end of last year to roughly $150 million by June. Legora itself had already confirmed crossing $100 million after less than 18 months of general availability.

Customer adoption is also moving quickly. Legora now says more than 100,000 legal professionals use the platform across more than 1,500 organizations. Earlier in the year it was serving about 1,000 organizations, while Financial Times reporting put the customer count around 1,200 during the summer.

Harvey has gone broader still. Recent reporting says more than 200,000 lawyers use the product, and Harvey has disclosed more than 25,000 custom agents built by its customers.

The gap therefore has two dimensions. Legora is multiplying revenue faster, while Harvey is creating more new revenue dollars and still has roughly twice the ARR.

Legal AI startup	Latest ARR	Earlier ARR	Approx. increase
Harvey	~$350M	~$195M	+~$155M
Legora	~$150M	~$50M	+~$100M

Why are AI coding agents growing faster than almost everything else?

AI coding agents are growing so quickly because developers are already using them every day, while many other enterprise-agent categories are still fighting their way through deployment.

A recent JetBrains survey of more than 15,000 professional developers found that 90% were using AI coding agents for work at least weekly and 68% were using them every day.

The deeper result is even more striking. Developers told JetBrains that agents fully generated about 47% of the code they produced on average. Roughly one in five respondents said they wrote zero code completely without AI help.

Those numbers describe a category that has already become part of normal work.

McKinsey's latest global AI survey points in the same direction from the enterprise side. Around 31% of respondents at companies with more than $1 billion in revenue said their organizations were scaling coding agents. Across all company sizes, nearly one-third said their employers had skipped buying at least one software product or feature because employees could build the functionality themselves with agentic coding tools.

That is a powerful growth loop for companies such as Cognition. Developers can begin using the product individually, teams can increase consumption quickly, and large enterprises can later roll the same tools across thousands of engineers.

Coding also gives agents something many office workflows lack: fast feedback. Code can be executed, tested, checked and repaired. An agent has much clearer evidence that its work succeeded.

The category is brutally competitive, with Claude Code, Codex, GitHub and Google all fighting for developers. Yet the size and frequency of software work currently give coding-agent startups one of the best environments in which to turn AI usage into enormous revenue.

Are companies really using AI agents, or are most deployments still pilots?

AI agents are moving into real production work now, although the explosive revenue of the leading startups is running well ahead of adoption across the average company.

LangChain surveyed more than 1,300 professionals and found that 57% had agents in production, up from 51% in its previous survey. Among organizations with more than 10,000 employees, the figure reached 67%.

McKinsey gives us a tougher benchmark because it asks about scaling across organizations rather than whether any production deployment exists. Its latest survey found that 40% of respondents from companies with more than $1 billion of annual revenue were scaling AI agents, up from 27% a year earlier. Smaller organizations remained around 22%.

The startup customer numbers fit between those two pictures.

Sierra says its agents already handle billions of interactions and are deployed at more than 40% of the Fortune 50. Cognition has reported more than tenfold growth in enterprise Devin usage. Legora has passed 100,000 legal professionals across more than 1,500 organizations. Harvey is now used by more than 200,000 lawyers, with customers building tens of thousands of their own agents.

Those deployments make it difficult to argue that agent revenue is still being driven mainly by demos.

At the same time, a company experimenting with one agent is very different from a company redesigning an entire operation around agents. The gap between those two stages remains huge. That helps explain how the leading vendors can grow several hundred percent while the broader enterprise market still feels early.

Can we really compare all these AI agent ARR numbers?

AI agent revenue figures are useful enough to rank companies, but they are nowhere near as standardized as traditional SaaS ARR.

Some companies report contracted annual recurring revenue. Others annualize their latest month of usage. “Annualized revenue,” “run-rate revenue” and “ARR” are often used almost interchangeably in private AI-company reporting even though they can describe different things.

Cognition's latest figure is a good example. The roughly $900 million number represents approximately $75 million of monthly revenue annualized. It does not mean Cognition collected $900 million during the previous twelve months.

Usage pricing creates another complication. If a large customer suddenly sends much more work through an agent, annualizing that month can make the business look as if its new level has already stabilized.

This is why we have relied on trajectories whenever possible.

Cognition has moved through several independently reported revenue levels rather than producing one isolated headline. Genspark passed $36 million shortly after launch, then $100 million and later $250 million. Sierra has moved through the $100 million, $150 million and $200 million range. Legora went from a few million dollars to $50 million, then beyond $100 million and roughly $150 million.

Repeated milestones give us much more confidence than one annualized month.

We should still avoid pretending that $100 million of Sierra ARR, $100 million of Decagon annualized revenue and $100 million of Manus run-rate revenue are accounting-identical. They are close enough to compare directionally, which is all this ranking really requires.

Is the fastest-growing AI agent startup also the best business?

No. The fastest-growing AI agent startup today can still have much worse economics than a slower software company.

Cognition shows the trade-off most clearly. The Information reports that the company could burn as much as $800 million in cash this year. Coding agents consume expensive models and computing infrastructure, and heavy users can generate enormous inference bills.

Wonderful has a different cost problem. Recent reporting puts its gross margin around 52%, far below what investors traditionally expect from mature SaaS. The company's large forward-deployed engineering operation helps customers install and adapt agents, but all that human work costs money.

Harvey is also growing quickly without being profitable yet. Legal AI requires expensive model usage alongside a sizable organization that helps major firms deploy the technology.

These economics do not invalidate the revenue growth. They change what that growth is worth.

If inference prices keep falling, companies route simpler work toward cheaper models and customers keep expanding usage, margins can improve sharply. Harvey has already begun using lower-cost models for some workloads where customers allow it.

For now, we would keep two rankings separate. Cognition leads the growth race. We still do not know which AI agent company will eventually turn this growth into the strongest long-term economics.

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

Which AI agent startup has the most durable growth?

Cognition has the strongest growth engine today, while Sierra and Harvey probably have the clearest evidence that large customers are embedding their agents deeply enough to stick around.

Cognition benefits from the sheer frequency of software work. Engineers can send tasks to coding agents every day, and the JetBrains adoption data suggests the behavior is rapidly becoming normal. The main risk is competition: Claude Code and Codex are gaining users fast, and model companies can subsidize or bundle their own coding products.

Sierra's position looks different. Its agents sit inside customer-service operations at major banks, insurers, telecom companies and other large enterprises. Once an agent is connected to company systems, trained around policies and handling real customer interactions, replacing it becomes a serious operational project.

Harvey has a similar advantage in legal work. Firms are putting internal knowledge, matter context and custom workflows into the platform. More than 25,000 customer-built agents suggest that usage is spreading beyond a single generic assistant.

Legora could become the biggest challenger in that vertical if its current growth continues. Its user count has risen much faster than its customer count, which suggests expansion inside existing organizations rather than growth coming only from new logos.

Manus remains the strongest general-purpose challenger to Cognition, but the next few quarters will be unusually revealing. The company has just gone through an acquisition, a regulatory reversal and a return to independence. Keeping its previous growth rate through all of that would make the business much harder to dismiss.

Genspark could also jump straight back into the conversation with one fresh revenue disclosure. Its last verified three-month acceleration was faster than Cognition's current percentage rate. We simply do not have enough new financial evidence to assume that pace continued.

Wonderful is the company most likely to surprise us. Seventyfold growth rarely survives as the revenue base gets larger, but even a major slowdown would leave the company growing unusually fast.

Which AI agent startup is growing the fastest?

Cognition is the fastest-growing independent AI agent startup as of now because no other company combines its current scale with a comparably fresh rate of expansion.

The strongest competing claims all require us to change the definition slightly.

Wonderful wins on percentage growth. Genspark produced the fastest verified three-month percentage burst after crossing $100 million ARR. Manus has multiplied a nine-figure revenue base several times in less than a year. Legora is growing faster than Harvey proportionally in legal AI. Sierra has built the largest specialist customer-service agent business in our comparison.

Cognition is the only company that currently sits near the top on both sides of the equation.

As seen above, its annualized revenue moved from roughly $492 million to more than $900 million in about one quarter. That means the company added around $400 million of annualized revenue while still growing at roughly 22% per month.

The Windsurf acquisition means we should avoid describing the entire rise as organic Devin growth. It does not explain the latest acceleration, which happened after the two businesses were already combined.

Manus is the closest challenger if we use a narrower definition centered on general-purpose autonomous agents. Wonderful is the obvious answer if someone asks only for the highest percentage growth. Neither currently matches Cognition's combination of size, recent speed and absolute revenue creation.

So if someone asks us today which AI agent startup is growing the fastest, Cognition is the clearest answer.

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

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.