Ema Raises $77M as Its Pitch to Enterprises Gets Blunter: Your Software Is Becoming a Database

Ema does not want to automate your enterprise software. It wants to make it irrelevant.

That is the blunt version of the pitch that just persuaded a group of investors to hand the three year old startup $77 million. The Series B, led by Bengaluru based Creaegis with Accel, Section 32 and Prosus all increasing their stakes, brings Ema’s total funding to $140 million and, according to the company, more than quadruples the valuation it carried after its last round in 2024. Ema would not put a number on the new figure, but the shape of the story is clear enough: a startup selling “AI employees” that do the work SaaS applications used to gate behind logins is now worth roughly four times what it was worth 18 months ago.

The round itself is unremarkable by 2026 standards, all primary equity, no debt, no secondaries cashing out early believers. What is less usual is the argument founder and CEO Surojit Chatterjee is making about why anyone should care, which is not really about Ema at all. It is about what happens to the rest of the software industry once agents like his get good enough.

The numbers behind the round

Ema was founded in 2023 by Surojit Chatterjee, previously chief product officer at Coinbase and a longtime Google executive, and Souvik Sen, who spent years building identity and access products at Okta. The company came out of stealth in March 2024 with a $25 million seed, added another $36 million four months later to push that round to $50 million, and had raised roughly $61 million total heading into this year. The new $77 million more than doubles that figure on its own.

Ema says it now serves more than 50 enterprise customers, including Google, Microsoft, PwC and KPMG, with over a million active users interacting with its agents. The company claims 50 fold revenue growth over two years and gross margins around 80%, along with a net dollar retention rate near 180%, meaning existing customers are expanding their contracts substantially rather than just renewing them. Those are self reported figures, worth treating the way any founder’s numbers deserve to be treated, but they are the kind of growth curve that explains why investors were willing to pay four times up from 18 months ago for a company that still has fewer than 250 employees.

What an “AI employee” is actually doing

Strip away the marketing language and Ema is an orchestration layer. The company sells a library of more than 150 prebuilt agents, each scoped to a narrow job, an HR agent that processes leave requests, an IT agent that resets access and provisions accounts, a finance agent that reconciles invoices, a compliance agent that checks a contract against a policy document. Customers can also build their own with a no code workflow editor. None of that is unusual on its own; plenty of vendors sell task specific bots.

The part Ema leans on hardest is the routing layer underneath, which it calls EmaFusion. Rather than betting on a single foundation model, EmaFusion breaks an incoming task into subtasks and picks, in real time, which of more than 100 available language models is best suited and cheapest for each piece, sometimes blending several outputs together to hit an accuracy target before returning an answer. It is the same instinct that is reshaping a lot of AI infrastructure spending right now, the recognition that no single model wins every job, which is part of why the market for routing and gateway products around models has gotten so much venture attention this year, including Stripe’s seven billion dollar acquisition of OpenRouter earlier this year to buy its way into exactly that layer.

Coordinating dozens of narrow agents behind one interface is also not a problem unique to Ema. It is close to the same architectural bet that shows up in developer tools now shipping coordinator agents that farm work out to thousands of disposable subagents, the pattern Cursor built for managing large coding projects across many parallel agents rather than one model trying to hold an entire codebase in its head. Ema’s version of that idea is aimed at HR tickets and vendor invoices instead of pull requests, but the underlying engineering problem, splitting work, routing it to the right specialist, checking the output, is the same one showing up across the agent economy this year.

The claim that makes this more than another funding round

What separates Ema’s pitch from a standard “we automate workflows” story is Chatterjee’s description of where this ends up. Speaking to reporters around the raise, he argued that once an agent layer sits on top of a company’s software stack and actually does the work, the underlying applications stop being products people use and start being storage. Ema, he said, tends to wrap around a customer’s existing systems first, and only later do customers begin peeling away the license fees for tools that Ema’s agents have made redundant. The applications do not disappear. They get demoted to holding the data while the agent layer does the deciding.

It is a convenient thing for the CEO of an agent company to say, and it should be read with that in mind. But it is not an isolated argument. Commentators covering enterprise software this year have started calling the broader pattern a “SaaSpocalypse,” shorthand for the idea that per seat software pricing built for a world of human operators does not survive contact with a workforce of agents that need no seats at all. Public SaaS multiples have wobbled on exactly this worry more than once in 2026, even before anyone could point to hard revenue numbers proving the software is actually being displaced rather than merely supplemented.

Set next to that table, Ema’s round looks modest, and that is worth sitting with for a second. A $77 million check landing at a fraction of Sierra’s or Decagon’s valuations is not evidence that investors doubt the thesis. It is evidence that Ema is earlier in its scaling curve than either of them, and that the category as a whole has enough capital chasing it that even the smaller checks now come with unicorn adjacent multiples attached. The same appetite has been showing up outside Silicon Valley too. India based enterprise AI startups have been pulling in outsized rounds this year on similar logic, including the $25 million raise by Deccan AI, a smaller but structurally similar bet that specialized, workflow specific AI talent and tooling is where enterprise budgets are actually moving.

Reasons to be skeptical

The gap between what founders say on funding day and what actually happens inside procurement departments is usually wide, and there are specific reasons to keep it wide here. Ema’s headline growth figures, the 50 fold revenue increase and the surpassing of $150 million in bookings, mix multiyear contract value with recognized revenue, a common practice in enterprise sales but one that makes year over year comparisons harder to pin down than the press release implies. A three year contract signed once counts toward bookings once, not annually, and inflates the apparent size of a young company’s business if read carelessly.

There is also a wider capital discipline question hanging over the entire category. Analysts at Goldman Sachs have described investor sentiment shifting this year from unconditional optimism about AI’s potential toward what they called a “show me the money” posture, and a meaningful share of early stage agent startups are believed to be burning through runway faster than expected, squeezed between high per query model costs and enterprise sales cycles that still move at the pace enterprises have always moved at, regardless of how fast the underlying technology improves. None of that is specific to Ema. All of it applies to Ema.

Where this leaves enterprise buyers

The most useful way to read this round is not as proof that SaaS is dying, which is a claim that has been made prematurely before, but as a marker of how confident the agent layer has gotten about its own position. Two years ago, tools like Ema sold themselves as assistants sitting next to your existing software stack, helping employees get through it faster. The pitch now is that the stack itself becomes optional once the agent layer is good enough, and investors just paid a four times markup to bet that the second version of that story is closer to true than the first.

Whether that bet pays off depends on something that funding rounds cannot settle on their own, which is what actually happens the next time a CFO’s contract renewal comes up for an application that an Ema agent has quietly been doing the real work around for a year. If that renewal gets smaller, or does not happen at all, Chatterjee’s line about software becoming a database stops being a talking point and starts being a balance sheet item for somebody else’s business. If it does not, the round still bought Ema a lot of runway to keep making the case.