AI agents that run a computer the way a person would - no APIs, no brittle scripts. Just show it the work.
Automat's mark, photographed against the machinery it aims to replace - the manual clicking and copy-pasting that quietly runs banks, insurers and lenders.
Most software wants an API. The real world runs on screens, PDFs and legacy portals nobody will ever integrate.
Automat, a roughly 25-person company headquartered at 660 York Street in San Francisco's Mission District, builds artificial-intelligence agents that operate a computer the way a human operator does - looking at the screen, reading the documents, filling the forms, and moving to the next task. It is a deceptively plain description for a hard problem. The manual, screen-based work that Automat targets is the connective tissue of the modern enterprise: a loan officer re-keying an appraisal, a compliance analyst running a KYC check across three government websites, a claims processor sorting a stack of medical PDFs. None of it is glamorous. All of it is expensive.
The founders' wager is that this work does not need another integration project or another army of contractors. It needs an agent that can see. Automat pairs a proprietary, transformer-based engine for robotic process automation (RPA) with intelligent document processing (IDP), then wraps both in a product designed so a non-engineer can build an automation by demonstrating it once. Show the agent a screen recording of the task, and it learns the steps.
In a field crowded with promises, Automat leads with receipts. The figures below come from the company's own reporting on production workloads - a reminder that in enterprise automation, reliability is the product.
For two decades, enterprises trying to automate repetitive computer work reached for tools like UiPath, Automation Anywhere and Blue Prism. These platforms record a fixed sequence of clicks and replay it. They work - until a web page redesigns, a field shifts position, or a vendor pushes an update. Then the script breaks, and a developer has to rebuild it. The maintenance burden is the dirty secret of legacy RPA.
Automat's pitch is that an AI-native agent removes that fragility. Because its agents reason about what is on the screen rather than following pixel-perfect coordinates, they can recover from errors and, the company says, self-heal when an interface changes. They are also built to run in the environments that give traditional automation the most trouble - Windows desktop applications, Citrix virtual sessions and messy legacy web portals. The result, per Automat, is deployment measured in days rather than months, with materially lower operating cost.
A UI-based AI agent builder for robotic process automation, powered by a transformer-based engine that deploys bots which mimic human behavior across Windows, Citrix and web portals.
Intelligent document processing that classifies and extracts data from unstructured PDFs, invoices, prescriptions and appraisals using fine-tuned vision and language models.
Create agentic workflows and document extractors by uploading reference materials - a screen recording or an SOP. Show the task; the agent learns it.
Forward-deployed engineering teams that build, deploy and maintain automations for enterprise clients who want outcomes rather than a toolkit.
Automat's customers cluster in the industries most burdened by document-heavy, regulated back-office work: banking and financial services, insurance, mortgage lending, healthcare and B2B software. The use cases are specific. Mortgage lender AmeriTrust uses Automat to automate disclosures, classify PDFs and extract appraisal data directly into its loan-origination system. An international payment processor runs KYC verification across government websites. A global insurance company relies on the platform to process millions of medical-claims documents - an effort the company says saves seven figures in operating expenses each year.
Other named users include the CPG revenue-management firm Vividly, which automates dispute filing across distributor websites, along with Ziina, Leopard, Ento and Whipsaw. It is unglamorous work by design. The value is not in a chatbot that writes emails; it is in the plumbing nobody wants to touch.
Lucas Ochoa and Gautam Bose met on the first day of design school at Carnegie Mellon University, both earning bachelor's degrees in design in 2019. They were hired together out of college into Google Creative Lab, Google's tech-and-culture incubator, where they spent three and a half years shipping products including the AI Test Kitchen, Teachable Machine and the Pixel Buds Pro. In 2023 they left to build their own company - first called Lasso, then renamed Automat as the ambition grew from lassoing individual workflows to operating the whole computer.
Previously Google Creative Lab and Google X. Leads company strategy and the enterprise go-to-market.
Previously Google Creative Lab, with work on transformer models tied to Google Chrome and GCP. Leads the agent and document-AI engineering.
Automat operates as a B2B enterprise software company with subscription and usage-based pricing, complemented by the managed-service model of Automat Workforce. That combination is a deliberate answer to the failure mode of legacy RPA, where customers buy licenses for a scripting tool and then quietly pay far more to keep the scripts alive. By packaging forward-deployed engineers alongside the platform, Automat sells the outcome - agents that stay live in production - rather than the raw software.
Within the market, Automat sits in the emerging "agentic automation" category, competing directly with the RPA incumbents it names - UiPath, Automation Anywhere and Blue Prism - as well as newer AI-native challengers such as Sola. Its differentiation rests on models that see and reason rather than replay, and on a design-led product that lets a citizen developer build automations by demonstration. The recognition is starting to follow: Automat was named IDP Startup of the Year by industry analyst firm Deep Analysis.
Ochoa and Bose graduate from Carnegie Mellon's School of Design, having met on their first day.
The pair are hired together, working on AI Test Kitchen, Teachable Machine and Pixel Buds Pro.
They leave Google, join Y Combinator's Winter 2023 batch, and raise a $3.75M seed from Initialized and Khosla.
Lasso becomes Automat as the mission expands from workflows to operating the whole computer.
Felicis leads a $15.5M Series A, bringing total funding to about $19.25M as deployments scale.
In November 2025, Automat closed a $15.5 million Series A led by Felicis, with participation from Initialized Capital, Khosla Ventures and Y Combinator, plus follow-on checks from K5 Global and Input Capital. Added to its earlier $3.75 million seed, that brings total capital raised to roughly $19.25 million.
Initialized Capital, Khosla Ventures and Y Combinator back the company in its earliest days as Lasso.
Led by Felicis, with Initialized, Khosla, Y Combinator, K5 Global and Input Capital participating.
Automat builds AI agents that operate computers like a human - handling click-based automation (RPA) and intelligent document processing (IDP) so enterprises can automate manual, screen-based workflows without heavy integration.
It was founded in 2023 by Lucas Ochoa (CEO) and Gautam Bose (CTO), Carnegie Mellon design alumni who previously worked together at Google Creative Lab. The company started as Lasso before rebranding to Automat.
Automat has raised about $19.25M total, including a $15.5M Series A led by Felicis in November 2025, with Initialized, Khosla Ventures, Y Combinator, K5 Global and Input Capital participating.
Unlike legacy RPA that relies on brittle scripts, Automat's AI-native agents see and reason about the screen, recover from errors, and self-heal when interfaces change - and can be built by showing the agent a screen recording rather than coding.
Enterprises in banking, financial services, insurance, mortgage lending and healthcare - including AmeriTrust, Vividly, an international payment processor running KYC, and a global insurer processing millions of medical-claims documents.
Official channels, coverage and the demo that started it all.