CopyCat Watches You Do Boring Work, Then Does It For You

The Macro: Back-Office Work Is Where Automation Goes to Die

Every company has back-office operations that run on human labor, spreadsheets, and prayer. Invoice processing. Claims adjudication. Order routing. Data entry from PDFs into systems that should talk to each other but don’t. It’s the kind of work that nobody glamorizes, nobody builds conference talks around, and nobody has successfully automated at scale despite thirty years of trying.

Traditional RPA (robotic process automation) was supposed to fix this. UiPath, Automation Anywhere, and Blue Prism built billion-dollar businesses on the promise of software robots clicking through enterprise applications the way humans do. The problem is that traditional RPA is brittle. It breaks when a button moves. It breaks when a form field changes. It requires dedicated developers to build and maintain. The average enterprise RPA implementation takes months and costs six figures before it automates a single task.

The AI wave has changed the underlying technology but not yet the delivery model. Most AI automation tools still require someone technical to set them up. You need to define the workflow, connect the APIs, handle the edge cases. That’s fine for a company with an engineering team. It’s useless for a 200-person healthcare company where the back-office runs on Karen and two other people who know how the system works.

What’s interesting right now is the convergence of browser agents, vision models, and API reverse-engineering. The technical building blocks exist to watch someone perform a task, understand what they’re doing, and replicate it. The question is whether anyone can package that into something reliable enough that a non-technical operations manager would trust it with real work.

The Micro: Show It Once, Let It Run

CopyCat’s pitch is straightforward: they analyze your manual processes and build custom AI automations to handle them. The toolkit includes browser agents (AI that can navigate web applications like a human), reverse-engineered API integrations (connecting to systems that don’t offer clean APIs), and document processing (extracting data from messy inputs like PDFs, faxes, and scanned forms).

The “agentic RPA” framing is deliberate. This isn’t the old model of recording a macro and replaying it. The agents are supposed to handle variation, adapt to edge cases, and keep working when the underlying applications change.

Graham Sabin and Abhi Balijepalli are two of the co-founders, with a third co-founder Zyad rounding out the team. Graham previously built Platter, a restaurant tech company processing hundreds of thousands of dollars monthly. Abhi built meeting infrastructure at Zoom. They’re based in San Francisco and came through YC’s Winter 2025 batch with Jared Friedman as their partner.

The backgrounds are relevant. Restaurant operations and video conferencing infrastructure are both domains where reliability at scale is non-negotiable. If Platter’s ordering system went down during dinner rush, restaurants lost money. If Zoom’s meeting infrastructure hiccupped, millions of people noticed. That operational DNA matters when you’re building software that’s supposed to replace a human team doing mission-critical work.

The target verticals appear to be healthcare, logistics, and general back-office operations. Healthcare is particularly interesting because the industry is drowning in manual processes that are too regulated to outsource casually but too tedious for skilled workers to spend time on. Prior authorization, claims processing, eligibility verification: all of it is manual labor at most organizations.

The competitive space includes Thoughtful AI (healthcare-specific RPA), Adept (general browser agents, now acquired), and the traditional RPA vendors who are bolting AI onto their existing platforms. CopyCat’s angle is the “custom” part. They’re not selling a product you configure. They’re analyzing your specific workflows and building agents tailored to them.

The Verdict

I like the positioning. “Replace your BPO or back-office team” is a bold claim, but it’s the right ambition for this space. The companies that are honest about wanting to replace headcount, rather than hiding behind phrases like “augmenting your workforce,” tend to build more focused products.

The risk is scope creep. Every customer’s back-office is different. If CopyCat ends up building bespoke automations for each client, they’ve built a services business with an AI wrapper, not a scalable software company. The magic trick is finding enough commonality across customers that the agents can be mostly reused while still handling the specific weirdness of each company’s operations.

At 30 days, I’d want to see how long it takes to go from first meeting to first automated task running in production. If it’s under two weeks, they have something. If it’s two months, they’re consulting. At 60 days, I’d want to see failure rates. How often do the agents break? How often does a human need to step in? At 90 days, the question is expansion within accounts. If one department automates successfully, do other departments start asking for it? That’s the signal that the product is working at a level where word spreads internally. The back-office automation market is massive and still mostly unaddressed. If CopyCat can actually deliver on the “show it once” promise, the customer acquisition motion almost runs itself.