Ema, the colleague
who never logs off.

The quietest hire of the year is a model parliament.

Walk into any enterprise call center built before 2024 and you can still feel the legacy of the rota. Shifts. Queues. Tier-one agents reading scripts that were written by tier-three managers who left the job in 2019. Ema's customers, almost all of whom run real businesses with real customers and real budgets, would prefer not to do this anymore. So they hire Ema instead.

Ema is not a chatbot. The founders are insistent on this point, sometimes a little wearily. It is, in their preferred terminology, a Universal AI Employee - a single platform that absorbs job descriptions the way a new hire absorbs a Notion onboarding doc, and then does the job. Customer support. Sales development. Compliance review. Proposal writing. Onboarding paperwork. Whatever the org chart calls it, Ema can probably be its newest member.

Two years out of stealth, the company has raised $61 million from Accel, Section 32, Prosus Ventures, Hitachi Ventures, Wipro Ventures, Sozo Ventures, SCB 10X, and - in a move that surprised even the seasoned skeptics - KPMG LLP. The customer roster has tripled. The team has tripled. The pitch deck has not changed.

Every enterprise wants AI. Almost none of them want to manage it.

The dirty secret of the generative AI boom is that the average Fortune 500 has roughly 14 LLM pilots running concurrently, no shared evaluation framework, three competing vector databases, and a Slack channel called #ai-governance that nobody reads. The C-suite knows productivity is on the table. The IT team knows the integration backlog is on fire. The board knows the budget is up for renewal in March.

Ema's founders looked at this and noticed something obvious that most of the industry was politely ignoring: enterprises do not want to buy AI. They want to hire it. They want the affordances of an employee - someone you can onboard, someone you can review, someone you can give credentials to, someone who shows up on Monday morning and knows what to do.

The bet is simple and quietly aggressive: that the era of buying tools is ending, and the era of hiring software is beginning. If Ema is right, the SaaS app is the last interface humans will ever click. If it is wrong, well - that is what the $61M is for.

One ran product at Coinbase. The other ran ML at Okta.

Surojit Chatterjee is the kind of founder VCs describe as "obvious" - which is not always a compliment, but in this case it might be. He was Chief Product Officer at Coinbase through its 2021 IPO. Before that, he scaled Google Mobile Ads and Google Shopping into multi-billion-dollar businesses. He holds 40 U.S. patents. He has an MBA from MIT, an MS from SUNY Buffalo, and a B.Tech from IIT Kharagpur. He left Coinbase in 2023 and gave himself one year to start a company.

His co-founder Souvik Sen led VP-level engineering for data, ML and devices at Okta. Before that he ran TrustGraph at Google - the machine learning system that flagged ad fraud across the open web. He is the kind of engineer who treats large-model orchestration the way a structural engineer treats a load-bearing wall: with respect, and with a calculator.

The two of them seem aware that "AI agents for the enterprise" is, in 2026, the most crowded pitch deck category in venture. They are equally aware that most of their competitors have shipped demos and not deployments. Ema's response is to release boring artifacts - SOC 2 reports, on-prem installers, KPMG integration playbooks - while everyone else releases launch videos.

EmaFusion is the bet. The Generative Workflow Engine is the receipt.

The technical architecture is the part of the pitch where most agentic AI companies handwave. Ema does not. Underneath the wordmark sits a system called EmaFusion - an orchestration layer that routes each query across more than 100 different models in real time, blends the outputs, and returns whichever answer scores highest against a domain-specific accuracy benchmark. Foundation models. Specialized models. In-house fine-tunes. It is, in effect, a small parliament of LLMs voting on every customer message.

Sitting on top of EmaFusion is the Generative Workflow Engine, or GWE. This is the part the buyer actually interacts with. GWE lets an enterprise describe a workflow in plain language - "when a refund request arrives, check the policy, verify the order, draft a reply, and route to a human if the amount exceeds $500" - and produces a running AI Employee that does exactly that. There are also 30+ pre-built employees, named like LinkedIn profiles: Agent Assist, Proposal Manager, Compliance Analyst, AI SDR.

Fig. 2 - Four product lines, one pitch. The on-prem option is the one most enterprises ask about first.

Numbers, on the record.

The case studies Ema is willing to publish involve real, named companies - which in this category is rarer than it should be. Envoy Global, TrueLayer, and Moneyview are all on the public roster. Internal benchmarks claim Ema performs at or above human baselines on the roles it has been hired into. The independent benchmarks will come; the cheque, in the meantime, is clearing.

Grow world GDP. Yes, really.

Ema's mission statement, in full, reads: "Grow world GDP by transforming every business with universal AI employees." It is the kind of phrase that, written on a slide, sounds either deeply earnest or deeply silly depending on the audience. The founders have been asked about it often enough that they no longer flinch.

The argument behind it is the unglamorous one: most knowledge work, in most industries, is a sequence of small, repetitive, badly-documented decisions. If you can automate those without losing accuracy, you free a measurable amount of human attention to do something else. Multiply that across 100,000 enterprises and you get a number that is, in fact, big enough to move a GDP line.

The interface is the last thing to die.

There is a version of the next decade in which every enterprise SaaS tool gets thinner, every workflow gets more autonomous, and the dashboard - that holy artifact of B2B software - finally retires. In that version, the unit of enterprise software is not the app. It is the employee. You hire it, you onboard it, you review it, and when it leaves, you offboard it.

Ema is not the only company betting on this future. Sierra is in the room. Decagon is in the room. Cresta and Sana and Glean are in the room. So are the in-house teams at Salesforce and ServiceNow, who have every incentive to win this category before the upstarts do. The competitive question for Ema is not whether the future arrives, but whether it arrives wearing the Ema wordmark.

The two-year head start on EmaFusion, the KPMG distribution channel, and a founder who has shipped at Coinbase and Google scale - these are the kinds of advantages that compound quietly until they don't. Whether they are enough is, in the end, the only question that matters.

Back to the insurance firm. The 2:14 a.m. ticket is closed. The customer never noticed. The morning shift will arrive at 9, read the log, and move on to the next thing. Somewhere in San Francisco, the company that closed the ticket is hiring its next ten engineers. The colleague who answered, true to her wordmark, is still online.