Automation Anywhere

Somewhere, a bot is reading your invoice.

It is a Tuesday morning at a global bank, and 4,000 invoices have already been processed before the espresso machine warmed up. Nobody clapped. Nobody filed a complaint. There is no employee of the month. The work just happened - read, extracted, validated, posted - by a quiet piece of software that the procurement team never quite learned to thank. That software, or a fair amount of it, came from Automation Anywhere.

From a tilt-roof office park in San Jose, this company has spent two decades building what its CEO Mihir Shukla likes to call a digital workforce. It is now used by more than 2,800 enterprises across 90+ countries. You may have never typed its name into a search bar. Your insurance claim, your onboarding form, and your refund request almost certainly have.

The most expensive employees doing the cheapest work.

Walk into the back office of any large enterprise and the picture is depressingly consistent. Smart people, with graduate degrees and good lighting, copy data from one screen into another screen. They reconcile spreadsheets. They click through 14-step approval flows authored by someone who left the company in 2011. The work is repetitive, unloved, and curiously immune to every promise tech companies have ever made about productivity.

The founders of Automation Anywhere noticed something almost embarrassing about this: it was not really a software problem. It was a translation problem. The systems that needed to talk to each other did, in fact, exist - they just refused to make eye contact. ERP did not speak to CRM. The mainframe did not speak to anything. The humans in the middle were the integration layer, and they were extremely bored.

Tethys Solutions, with a different name and a worse logo.

In 2003, Mihir Shukla, Neeti Mehta Shukla, Ankur Kothari, and Rushabh Parmani started a company called Tethys Solutions, named after the Greek goddess of fresh water - a fact you will never need at a dinner party but will probably enjoy anyway. They built a desktop macro recorder. It was not glamorous. It was, by the polite standards of Silicon Valley, almost charmingly unfashionable.

Then they did the thing the Valley does not encourage. They bootstrapped. For roughly a decade. They sold software to actual customers, charged actual money, and used the revenue to hire engineers rather than recruiters. By the time they rebranded as Automation Anywhere in 2010, they already had pilots reporting 30 to 60 percent time savings on targeted processes - the sort of numbers that make a CFO put down their coffee.

One platform, several increasingly intelligent residents.

The flagship today is Automation 360, a cloud-native platform that has absorbed every fashionable acronym of the last ten years without losing its plot. RPA lives there. IDP lives there. Process discovery lives there. AI agents have just moved in upstairs.

A 20-year sprint, told in receipts.

Customers, partners, and a marketplace that thinks it is an App Store.

The customer roster reads like a long lunch with the global economy. Banks. Insurers. Telcos. Hospital systems. Government agencies. The kind of organizations that do not make impulse purchases. They run Automation 360 against the work that used to require three open browser tabs and a quiet sigh.

Behind them sits an unusually deep partner stack. Google Cloud for Vertex AI. AWS, where Automation 360 ships on the Marketplace and integrates with Bedrock. Microsoft, for Azure OpenAI. Salesforce, where automations live inside Service Cloud. Workday, for HR and finance flows. The Bot Store, meanwhile, has thousands of pre-built bots - a marketplace where developers can publish and monetize automations, which is the part of the story that quietly resembles a platform company more than a software vendor.

Take the robot out of the human.

The official mission is to "enable people with barrier-free automation technology that liberates everyone to achieve their human potential." That is the press release version. The internal version, which Shukla has been repeating for years, is simpler: take the robot out of the human. Give people back the cognitive bandwidth they have been spending on data entry, and trust them to spend it on something better.

It is the kind of mission that sounds soft until you do the math. If a credit operations team gets twelve hours back per analyst per week, that is twelve hours someone is not staring at a screen wondering whose idea this all was. Multiply by 2,800 enterprises. Multiply by 90 countries. The unit economics, eventually, become a question about how humans want to spend the workday.

Agentic AI was always going to need a chaperone.

The current wave of agentic AI is built on a quiet assumption: that autonomous software can be trusted to do real work inside real companies. That assumption requires governance, observability, integration with messy legacy systems, and a way to put a kill switch on a model when it gets too creative. Almost none of that is solved by an LLM API. It is solved, awkwardly and unglamorously, by the kind of orchestration layer Automation Anywhere has been building since before "agentic" was a word anyone repeated in a board meeting.

Which is why the AI Agent Studio matters more than its product page suggests. It is not just a builder for agents. It is the place where the enterprise gets to say: yes, this agent can talk to SAP; no, this agent cannot touch payroll; and here, in audit-friendly logs, is everything it did last Tuesday.

The invoice that nobody noticed.

It is still Tuesday. The bank has now processed 6,200 invoices. The procurement team has gone to lunch. Somewhere on a hallway whiteboard, someone has written the words "AI strategy" in green marker, underlined twice. The actual AI strategy, the one already in production, has been running quietly behind a login screen for years. It does not have a mascot. It does not appear in the all-hands deck. It just keeps clearing the queue.

This is what Automation Anywhere actually built. Not the future of work as a slogan. The present of work as a fact. The robot, taken out of the human. The human, presumably, doing something more interesting with the morning. The invoice, paid.

If you would rather hear it.