What are the top agentic AI startups by revenue today?

The ranking of top startups in the agentic AI market by revenue

Below is a table ranking all the companies in this market by their current revenue scale. You can find our methodology at the end of this page.

If you want a deeper understanding of the market and its current dynamics, get our report on agentic AI funding, valuations and who's ahead.

What are the top agentic AI startups by revenue today?

Anthropic's Claude Code, Cognition, Genspark, EliseAI, Sierra and Replit are currently the biggest agentic AI startup businesses for which we found strong public revenue evidence.

Claude Code is in a league of its own at more than $2.5 billion in revenue run-rate. Cognition follows at $900 million. After that, the numbers drop sharply: Genspark says it has reached $250 million ARR, EliseAI reports $200 million, Sierra is above $150 million and Replit reported $150 million in annualized revenue.

We exclude Salesforce Agentforce and Zendesk AI Agents from the main leaderboard because the question is about startups. Their agent products matter when we look at the wider market, but putting Salesforce beside Cognition or Sierra would distort what we are trying to measure.

The ranking also keeps ARR, annualized revenue and revenue run-rate separate. Private AI companies rarely hand us neat audited annual revenue figures, and pretending those metrics are identical would make the table look cleaner than the evidence really is.

Rank	Startup	Best current revenue evidence	Metric	Main agent market	Evidence
1	Anthropic / Claude Code	>$2.5B	Product revenue run-rate	Coding agents	Company disclosed
2	Cognition	$900M	Annualized revenue run-rate	Coding agents	Company disclosed
3	Genspark	$250M	ARR	General work agents	Company disclosed
4	EliseAI	$200M	ARR	Housing & healthcare agents	Company disclosed
5	Sierra	>$150M	ARR	Customer-experience agents	Company disclosed
6	Replit	$150M	Annualized revenue	App-building agents	Company disclosed
7	Manus	>$125M	Revenue run-rate	General-purpose agents	Company disclosed
8	Capacity	>$100M	ARR	Support automation agents	Company disclosed
9	Emergent	$100M	ARR	App-building agents	Company disclosed
10	Decagon	~$100M	Estimated annualized revenue	Customer-service agents	Third-party estimate
11	Wonderful	~$70M	Estimated revenue run-rate	Enterprise agents	Tier-1 reported estimate
12	Retell AI	~$60M	ARR	Voice agents	Company-linked disclosure
13	Rogo	~$53M	Estimated ARR	Financial-services agents	Third-party estimate
14	Parloa	>$50M	ARR	Customer-service agents	Company disclosed
15	PolyAI	~£17.9M	Filed turnover	Voice agents	Filed accounts

Does Anthropic really belong in an agentic AI startup ranking?

Yes, but Claude Code is the number we should use rather than Anthropic's total company revenue.

Anthropic sells much more than agents. Claude subscriptions, API usage and enterprise products all contribute to the wider business, so putting the whole company into an agentic AI ranking would exaggerate the size of its agent business.

Claude Code gives us a much cleaner perimeter. Anthropic now describes Claude Code as an agentic coding product with more than $2.5 billion in run-rate revenue. The company also says enterprise usage represents more than half of Claude Code revenue and that business subscriptions have quadrupled since the start of the year.

Claude Code is unusually easy to isolate compared with most agent products inside larger AI companies. We can see a real agent-specific commercial figure instead of borrowing revenue from unrelated products.

How big is Cognition right now?

Cognition is currently the clear number two, with a reported $900 million annualized revenue run-rate.

That figure was $492 million only a few months earlier. In other words, Cognition added more than $400 million of annualized revenue scale in a very short period.

The size gap is more useful than the growth percentage. Cognition's $900 million run-rate is 3.6 times Genspark's $250 million ARR and six times Sierra's reported $150 million-plus ARR.

There is one important limitation: $900 million is a run-rate, not $900 million of audited revenue earned over the previous twelve months. Cognition has not publicly explained the calculation in detail. Still, the gap is huge.

Among independent agent startups, Cognition is currently operating at a scale that almost nobody else has reached.

Is Genspark really making $250 million a year?

Genspark's $250 million ARR claim now has enough supporting evidence to take seriously, although it remains a company-reported annualized figure rather than audited revenue.

The trajectory is unusually well documented. Anthropic's earlier Genspark case study put the company at $36 million ARR only 45 days after launching Super Agent. The newer Claude customer material says Genspark has since surpassed $250 million ARR after pivoting to the Super Agent model. Going from $36 million to $250 million shows how quickly things move; we cover what changed in the past year in our report on agentic AI.

That is roughly seven times the earlier $36 million milestone.

Genspark is also a useful case because the revenue jump followed a clear product change. The company moved away from being mainly an AI search product and built an agentic workspace capable of orchestrating more than 150 tools for research, documents, slides, spreadsheets and other work.

The current $250 million figure puts Genspark comfortably behind Cognition but well ahead of the large cluster around $100 million to $200 million.

Which agentic AI startups have already crossed $100 million?

At least nine independent agentic AI startups now have credible public evidence around or above $100 million in ARR, annualized revenue or revenue run-rate.

The list stretches across several very different businesses. Cognition and Replit focus on software creation. Genspark and Manus are broader work agents. EliseAI automates housing and healthcare operations. Sierra, Capacity and Decagon concentrate heavily on customer experience. Emergent lets people build applications through natural-language interaction.

The breadth is more interesting than the exact count. A year ago it was reasonable to wonder whether agents would become a real software category. Now we have multiple $100 million businesses across several agent categories.

Startup	Current scale	Metric	Agent category	Evidence
Anthropic / Claude Code	>$2.5B	Revenue run-rate	Coding	Direct
Cognition	$900M	Revenue run-rate	Coding	Direct
Genspark	$250M	ARR	General work	Direct
EliseAI	$200M	ARR	Vertical operations	Direct
Sierra	>$150M	ARR	Customer experience	Direct
Replit	$150M	Annualized revenue	App building	Direct
Manus	>$125M	Revenue run-rate	General purpose	Direct
Capacity	>$100M	ARR	Support automation	Direct
Emergent	$100M	ARR	App building	Direct
Decagon	~$100M	Annualized revenue	Customer service	Estimated

Are coding agents making more money than customer-service agents?

Coding agents currently own the biggest revenue numbers, while customer-service agents have the deeper bench.

Claude Code above $2.5 billion and Cognition at $900 million create two enormous outliers. Replit adds another $150 million in annualized revenue, and Emergent has reached $100 million ARR.

Customer service looks different. Sierra is above $150 million, Capacity has crossed $100 million, Decagon is estimated around $100 million, Wonderful is near a $70 million run-rate, Retell is around $60 million and Parloa is above $50 million.

So coding is producing the giants for now. Customer service has more companies reaching serious scale.

The commercial logic is pretty straightforward. Coding agents can tap directly into highly paid engineering work and, increasingly, software creation by non-developers. Customer service has a broader pool of repetitive enterprise work, which creates room for more vendors.

Why are coding agents growing so fast?

Coding agents are growing fast because customers can see the output and put a value on it almost immediately.

A software agent either builds the feature, fixes the bug or produces the application. That is easier to measure than a broad promise that an AI assistant makes knowledge workers "more productive."

The category has also expanded well beyond autocomplete. Cognition's Devin is designed for longer software-engineering tasks. Replit increasingly sells the ability to create working applications from natural language. Emergent says around 70% of its users had never written a line of code before using the platform. Plain-language app building is a technology bet worth watching; we follow what is gaining traction in our agentic AI report.

That last point changes the size of the market. These products are starting to sell software creation to people who were never traditional developer-tool buyers.

How fast are the biggest AI agent startups growing?

The fastest agent startups are still adding revenue at a pace that looks extreme even by SaaS standards.

Cognition moved from $492 million to $900 million in annualized run-rate revenue within a few months. Genspark went from the $36 million ARR milestone highlighted in Anthropic's earlier case study to $250 million. Emergent says ARR doubled from $50 million to $100 million in a single month.

Sierra reached $100 million ARR in seven quarters and entered its third year above $150 million. Capacity says it went from $5 million to more than $100 million ARR in three and a half years. EliseAI reports five consecutive years of 100% year-over-year ARR growth.

These growth rates will not continue indefinitely. They do explain why a ranking like this moves much faster than a normal enterprise-software leaderboard.

Startup	Earlier revenue evidence	Current evidence	What changed
Cognition	$492M run-rate	$900M run-rate	Nearly doubled within months
Genspark	$36M ARR	$250M ARR	Roughly 7x
Emergent	$50M ARR	$100M ARR	Doubled in one month
Sierra	$100M ARR	>$150M ARR	Added >$50M
Capacity	$5M ARR	>$100M ARR	Roughly 20x in 3.5 years
EliseAI	Prior-year base roughly half today's level	$200M ARR	Fifth straight year of 100% growth
Manus	$100M ARR	>$125M total run-rate	Usage revenue pushes total above ARR

Why does an agentic AI revenue ranking use so much ARR and run-rate?

ARR and run-rate dominate this market because most of the fastest-growing agent companies are private and changing scale much faster than annual accounts can capture.

Cognition illustrates the problem. A completed fiscal year could be technically cleaner while being a poor description of a business whose current revenue pace has changed by hundreds of millions of dollars in a few months.

We still keep the labels separate. ARR remains ARR. Annualized revenue remains annualized revenue. A run-rate remains a run-rate. We do not turn one into another just to make the ranking easier to scan.

That distinction becomes especially important when usage-based revenue is growing quickly. Manus, for example, reports $100 million ARR but more than $125 million in total revenue run-rate once other usage revenue is included.

Which agentic AI startups have actual filed revenue?

PolyAI gives us one of the cleaner conventional revenue figures because its UK accounts are publicly filed.

The latest group accounts cover the year ending in January and report roughly £17.9 million in turnover. That number has a much stronger accounting basis than a founder announcing an ARR milestone on social media.

Filed revenue does not automatically deserve a higher ranking, though. PolyAI's accounting number describes a completed period, while businesses such as Cognition can change dramatically within a single quarter.

We therefore care about both evidence quality and how well the figure represents the business now. A strong filed number wins when two datapoints are otherwise similar. It should not automatically outrank a fresh direct disclosure that is many times larger.

Is EliseAI really an agentic AI company?

EliseAI belongs in the agentic AI market because its software now handles operational workflows across housing and healthcare rather than simply answering questions.

The $200 million ARR figure is especially interesting because EliseAI predates the latest agent boom. The company says it has doubled ARR every year for five straight years. Doubling ARR five years running points to demand worth finding; we look at where the biggest opportunities are in our report on agentic AI.

Housing agents can handle leasing and resident communications, while the healthcare business addresses similarly repetitive operational work. Those workflows involve integrations, state-level rules and a large number of messy edge cases.

EliseAI has a different profile from the latest generation of general-purpose agents. Its growth comes from going deep inside a few expensive workflows rather than trying to become an AI worker for everything.

Are customer-service AI agents already a real software category?

Customer-service agents have clearly become a real software category, with several independent companies now above $50 million in ARR or revenue run-rate.

Sierra is above $150 million ARR. Capacity has passed $100 million. Decagon is estimated around $100 million. Wonderful has reached an estimated $70 million annual revenue run-rate. Retell is around $60 million ARR and Parloa has surpassed $50 million.

The products are also doing more than handling basic FAQ chats. Sierra sells outcome-based agents, Decagon agents can take actions inside backend systems, Parloa focuses on voice and digital customer interactions, and Retell's agents can book appointments, qualify leads and complete support workflows.

This category is becoming crowded fast. The interesting question now is which vendors can keep a real advantage once good conversational quality becomes normal.

How big are voice AI agent startups now?

Voice agents are already producing $50 million-plus businesses, although the category is still smaller than the biggest coding-agent companies.

Retell is one of the clearest examples. OpenAI reported that Retell had generated $14 million in revenue within 16 months of launch. Retell subsequently disclosed $40 million ARR, then $50 million, and current company-linked material puts the business around $60 million ARR.

Parloa has passed $50 million ARR with enterprise voice and customer-service agents. PolyAI's filed group accounts show roughly £17.9 million in turnover.

Usage is even larger than the revenue figures suggest. Retell says it handles tens of millions of real-time calls every month. Other voice platforms such as Vapi disclose huge call volumes without giving us a reliable dollar revenue number.

Voice AI has moved well beyond demos, but revenue transparency remains uneven.

Is Manus still one of the biggest general-purpose AI agent startups?

Manus remains one of the largest general-purpose agent startups we can rank with a direct financial disclosure.

The company reported $100 million ARR and more than $125 million in total revenue run-rate once usage-based and other revenue were included. That is enough to keep Manus firmly in the top ten. Revenue is only one side of the story; we also cover what each company was worth then and now in our agentic AI market report.

Manus also reported more than 20% month-over-month growth after the release of Manus 1.5. We do not compound that growth rate into a new estimate because even a small slowdown would make the resulting number wrong very quickly.

For now, the last hard figure is already strong enough: Manus is a $100 million-plus agent business without any need for synthetic extrapolation.

Are general-purpose AI agents actually making serious money?

Yes, but the money is concentrated in a surprisingly small number of general-purpose agent companies.

Genspark at $250 million ARR and Manus above a $125 million total revenue run-rate show that broad agents can become large businesses. Below them, public dollar evidence gets thinner quite quickly.

Relevance AI, for example, reports substantial agent-task volume without an absolute revenue disclosure. Many other horizontal platforms talk about users, tasks, agents deployed or workflow volume.

"General-purpose agent" is too broad to explain why people pay. Genspark packages broad capabilities into a workspace that produces concrete outputs. Manus sells task execution and usage.

Broad capability can attract attention. Clear paid jobs are what turn that attention into revenue.

Are vertical AI agents getting as big as general-purpose agents?

Vertical agents can already get very large, and EliseAI is the strongest example in this ranking.

EliseAI's $200 million ARR puts a housing-and-healthcare specialist ahead of nearly every general-purpose agent startup except Genspark and the much larger coding businesses.

Rogo shows the same pattern lower down the ranking. Sacra estimates roughly $53 million ARR for its financial-services platform, which is designed around investment banking and institutional workflows rather than generic knowledge work.

Vertical companies can charge for expertise that is difficult to reproduce with a general agent: regulated workflows, proprietary data, integrations, permissions and industry-specific processes.

General agents have a much larger theoretical market. Vertical agents often have a clearer reason for a customer to sign a large contract today.

Does raising more money tell us which AI agent startups make the most revenue?

Funding is a poor shortcut for revenue in agentic AI.

Some highly funded agent startups still do not disclose an absolute revenue figure, while companies such as Capacity and Retell have reached large ARR milestones with much less capital than the best-funded names. Funding and revenue do not always line up; we look at who has raised the most, and from whom, in our report covering agentic AI.

Factory is a good example of why funding cannot substitute for revenue. The company has raised hundreds of millions of dollars and says hundreds of thousands of developers use its Droids. It also reported doubling revenue month over month for six consecutive months. Without the starting revenue base, we still cannot put a defensible dollar figure into the ranking.

Valuation has the same problem. Investors can price future growth years before the revenue becomes visible.

When the question is who makes the most money now, financing belongs in the background.

Which important AI agent startups still do not disclose revenue?

Factory, Vapi, Hippocratic AI, Assort Health and several other large agent companies remain impossible to place precisely because their strongest public figures are usage or growth metrics rather than absolute revenue.

Factory says its Droids are used every day by hundreds of thousands of developers. Hippocratic AI reports more than 250 million patient interactions. Assort Health says its platform has handled more than 190 million patient interactions and that revenue increased 20-fold over 15 months.

Those are substantial commercial signals, but none tells us exactly how many dollars the company generates.

This is where bad rankings usually create fake precision. They multiply users by prices, calls by estimated fees or valuations by guessed revenue multiples. We leave the revenue unknown instead.

Startup	Strongest public commercial evidence	Agent category	Why we do not revenue-rank it
Factory	Hundreds of thousands of daily developers; repeated revenue doubling	Coding agents	Absolute revenue base undisclosed
Vapi	Very large voice-call volume	Voice infrastructure	Dollar ARR undisclosed
Hippocratic AI	250M+ patient interactions	Healthcare agents	Revenue undisclosed
Assort Health	190M+ patient interactions; 20x revenue growth	Healthcare agents	Starting revenue base undisclosed
7AI	Millions of autonomous security investigations	Cybersecurity agents	Revenue undisclosed
XBOW	100+ customers	Cybersecurity agents	Customer count is not revenue
Relevance AI	Large monthly agent-task volume	General agent platform	Usage cannot be converted safely
Basis	Used by a large share of major accounting firms	Accounting agents	Customer penetration is not revenue

Which agent startups are probably underrated by this ranking?

Companies with old dollar figures but much fresher operating evidence are the most likely to look artificially small.

HappyRobot is a good example. Its available absolute ARR figure is much older than later evidence showing a growing enterprise deployment base. Bland AI has an even older revenue milestone alongside much larger recent call volumes.

We could create updated estimates for both companies, but doing so would mostly measure our assumptions.

A conservative ranking therefore rewards companies that disclose financial metrics frequently. We accept that bias because it is better than pretending usage can be translated neatly into revenue.

How much should we trust estimated AI startup revenue?

Estimated revenue is useful when the source is credible and the estimate fills a real gap, but we give direct company disclosures more weight.

Decagon is the clearest case near the top. Sacra estimates the company reached roughly $100 million in annualized revenue. The figure fits the company's rapid enterprise expansion, but it is still an external estimate. Companies that rely on outside estimates leave real gaps, and we go through the biggest unsolved problems in our agentic AI report.

Rogo's roughly $53 million ARR figure comes from the same type of research. Wonderful's estimated $70 million annual revenue run-rate was reported by CTech.

Small differences between these figures should not be taken too literally. A directly disclosed $50 million-plus ARR can be stronger evidence than a $53 million estimate even if the estimated number looks slightly larger.

Precision should come from the evidence, not from the number of digits on the page.

Can older AI companies still count as agentic AI startups?

Yes, as long as they remain independent growth companies and agents are genuinely central to what they sell now.

Capacity is a good example. The company existed before the current agent wave, but its present product is built around agentic support automation, and it still behaves much more like a private growth company than an incumbent software giant.

EliseAI fits for similar reasons.

Salesforce does not. Agentforce may be one of the world's largest agent products, but Salesforce is a mature public software company. Zendesk presents the same problem.

Keeping that boundary makes the ranking much easier to understand. We are comparing companies built around the current AI-agent opportunity, rather than every software vendor that has added agents to its product portfolio.

Is agentic AI still mostly hype?

The commercial evidence is now too large to describe agentic AI as mostly hype.

We found multiple independent startups above $100 million in ARR, annualized revenue or run-rate. Several more are already between $50 million and $100 million. The revenue is spread across coding, customer service, app building, voice, healthcare, housing and financial services.

The hype sits elsewhere. Funding rounds, valuations and impressive demos still get mixed together with actual revenue far too often.

A company with a billion-dollar valuation can have less commercial scale than a quieter company with $100 million ARR. A platform handling hundreds of millions of agent interactions can still have undisclosed revenue.

The market has become real. The hard part now is measuring it without confusing attention, usage, valuation and revenue.

What do the biggest AI agent startups actually sell?

The biggest agent startups sell completed work that customers already know how to value.

Claude Code and Cognition sell software-engineering output. Replit and Emergent let people create functioning applications. Sierra, Capacity, Decagon, Retell and Parloa handle customer interactions and support work. EliseAI automates housing and healthcare operations. Rogo helps financial professionals complete research and deal workflows.

That common thread explains more than the word "agent." Selling work that already has a price is one way to earn, and we compare which business models are winning in our report covering agentic AI.

The products at the top are attached to expensive work with a clear before-and-after result. Code gets shipped. An application gets built. A support issue gets resolved. A call gets handled. A leasing workflow gets completed.

Autonomy becomes much easier to monetize when the output already has a price.

How quickly can this agentic AI revenue ranking change?

This ranking can change within months because several leaders are still adding tens or hundreds of millions of annualized revenue at extraordinary speed.

Cognition's move from $492 million to $900 million is the clearest recent example. Genspark's climb from tens of millions to $250 million ARR also happened in a very compressed period. Emergent says it doubled from $50 million to $100 million ARR in one month.

Small gaps near $50 million to $100 million are especially fragile. A company ranked twelfth today could move several positions after one new disclosure.

The larger gaps are more meaningful. Claude Code and Cognition currently form one revenue tier. Genspark, EliseAI, Sierra and Replit form the next. A dense group then begins around $100 million to $125 million.

So who is actually winning the agentic AI revenue race right now?

Claude Code is currently the biggest agentic AI startup product by disclosed commercial scale, Cognition is the strongest independent pure-play challenger, and the next major group is Genspark, EliseAI, Sierra and Replit.

Claude Code's more than $2.5 billion run-rate is far ahead of the market. Cognition's $900 million figure is also several times larger than the companies below it. Genspark sits around $250 million ARR, EliseAI at $200 million, Sierra above $150 million and Replit at $150 million annualized revenue.

Manus follows above $125 million. Capacity and Emergent have crossed $100 million, while Decagon is estimated around that level. Wonderful, Retell, Rogo and Parloa make up the next visible $50 million to $70 million group.

The bigger picture is clearer now than it was even recently. Coding agents currently produce the giant revenue outliers. Customer service has the largest cluster of scaled startups. Vertical agents such as EliseAI show that specialization can produce very large businesses, while general-purpose agents only stand out financially when they are attached to concrete work people will pay for.