autotab

The interesting thing about Autotab isn't the AI label, it's the teaching interface. Most browser agents make you either write code or assemble a visual flow chart of steps. Autotab asks for a video of you doing the job, or a written document of the steps, and works from that. Anyone who has onboarded an offshore ops team will recognise the shape: you demonstrate once, you explain as you go, then it should run unattended. That framing matters because the workflows in the vendor's own use-case list are exactly the ones that resist tidy connectors — government forms that live behind a portal with an awkward upload step, vendor RMS/TMS screens, refund flows in internal admin apps. What separates it from a thin wrapper is the documented primitive set. Loops let it churn through a list of records. Conditionals let it branch when the screen differs from expectation. The scratchpad holds values between steps so state survives a page transition. The Agent step is where reasoning is allowed to happen inside an otherwise deterministic script, and Smart Click exists because real enterprise UI has elements that ordinary selectors miss. File handling and Google Sheets close the loop so data doesn't dead-end. That is an engineering surface, not a prompt box. The honest constraint is the Desktop App. You teach workflows from a local install, there is no purely web-based way to train a new skill, and the agent executes in its own secure local browser. For a security team that's a feature — credentials and session live on your machine, not a vendor's multi-tenant runner. For a fully remote, Chromebook-only ops team, or an organisation that wants bots running on a headless server farm, it's a real architectural question you should settle before signing anything. The vendor also won't promise universal coverage: they ask you to send a task recording and will tell you within 24 hours whether the agent can handle it. Use that. It's free and it is the fastest way to find out whether your specific portal is in scope. Where it doesn't fit: single-step automations, anything already covered by a native integration between two mainstream SaaS products, and processes that only fire a few times a month. The build-a-teaching-video cost per workflow only pays back when the process repeats often. The competitive frame is straightforward. Against legacy RPA, Autotab's pitch is that you teach a task instead of modelling every step in a designer, and that reasoning is available when a screen changes. Against BPO, the pitch is cost and 24/7 running. Against Zapier, the pitch is the entire long tail of web apps that will never ship a public API. If your process is in that long tail and happens daily, this is one of the few credible options.