Hunch vs RPA and browser automation

If Hunch isn't a chatbot, what is it? The closest existing category is robotic process automation (RPA) and browser-automation tools. This guide compares the two approaches honestly and explains when each makes sense.

What RPA and browser-automation tools do

Tools like UiPath, Microsoft Power Automate, Bardeen, and Magical automate repetitive browser work by recording and replaying user interactions.

The common model:

1. You record a workflow once, either by performing the steps or by pointing at UI elements.
2. The tool stores the workflow as a sequence of selectors — typically CSS selectors, XPath expressions, or screen coordinates.
3. A robot (desktop-attached, cloud-hosted, or browser-extension) replays that sequence on demand or on a schedule.
4. When the workflow ends, you wire the result into a downstream step — email, spreadsheet, database, or another system.

RPA is a mature category with enterprise governance around it: process documentation, change management, and audit trails. That maturity is real, and it solves a different set of problems than Hunch does.

Why RPA workflows break

The replay model is the weakness. A recorded workflow is a fragile artifact:

• Selector fragility. The workflow depends on specific CSS classes, XPath paths, or DOM structure. One class name change, one reordered form section, or one A/B test and the selector stops matching. Failures are usually silent until someone notices the workflow hasn't run.
• Layout fragility. Coordinate-based automation breaks on any viewport or design change. A responsive layout that shifts elements between desktop and mobile renders is a constant source of drift.
• Per-machine maintenance. Desktop robots are installed per machine or per VM. Each installation has its own version, its own browser profile, and its own failure modes. When the environment changes, the workflow breaks for that machine only.
• No learning. Every workflow is a bespoke recording. Nothing a robot learns on one form carries over to the next. A company with 40 forms maintains 40 recordings, and each form change is a maintenance task.
• No notion of intent. The robot knows where to click, not why. It cannot decide between filling a field and answering a question, and it cannot adapt when the form is replaced with an entirely different flow.

How Hunch differs

Hunch takes a different approach: instead of recording selectors, it understands forms at the field level and executes actions through the page's own DOM.

DOM-native execution resolves fields semantically

Hunch resolves form fields by name, id, aria-label, visible label text, and placeholder — not by CSS class or coordinates. When a label changes but the name attribute stays the same, execution keeps working. When a React or Vue component re-renders, Hunch sets values in a way the framework recognizes, then dispatches native submit events.

Four-layer dispatch, not record-and-replay

Hunch has four independent layers for executing an action:

1. Owner-declared manifests — a site operator publishes a hunch-actions.json describing actions and parameters. Highest trust, because the owner defined the contract.
2. Platform official APIs — for known platforms (for example, Shopify Storefront or WooCommerce REST), Hunch can use the platform's documented API instead of touching the DOM at all.
3. Passive live capture — Hunch observes real fetch/XHR traffic on pages where it runs, learning the shape of endpoints that already work in production. Privacy-preserving: only structure is stored, never request/response values.
4. DOM-native execution — the default: find the form, fill it, submit it.

Each layer is independent, not a fallback chain. The best available layer wins for a given action.

An approval gate on mutations

Anything that writes — POST, PUT, DELETE, anything touching money, accounts, or data — requires explicit user confirmation in the widget before it executes. RPA robots typically fire and forget; Hunch treats every state-changing action as something the visitor approves first.

Self-registering modals

Forms that live in modals or SPA components register themselves with Hunch's tool registry when they mount and unregister when they unmount. Hunch reads the registry synchronously — no DOM scanning, no polling, no guessing about which trigger opens which modal.

A network hub that learns across sites

Hunch shares endpoint discoveries, form patterns, and execution feedback across sites running Hunch in the same workspace (with per-website rate limiting and confidence gating). A form pattern learned on one site can inform coverage on another — something a per-machine recording can never do. The more sites run Hunch, the better the coverage on each one.

Honest scope

Hunch operates forms on:

• Pages where you have installed the Hunch widget
• Sites running Hunch
• Sites with an owner-published action manifest

It is not a tool for pointing at arbitrary third-party websites and automating them without the owner's involvement. If that is the requirement, an RPA tool is the honest answer.

When RPA still makes sense

RPA is the right tool when your problem is not a web form:

• Desktop applications — native Windows/macOS apps, legacy terminals, and mainframes that have no DOM and no API. Hunch operates web pages; it does not drive native apps.
• Cross-application orchestration — workflows that stitch together an ERP, a spreadsheet, a mail client, and a web portal in one sequence.
• Unattended batch processing — thousands of identical transactions run overnight, where a human approving each one defeats the purpose.
• Strict enterprise governance — environments that require formal process documentation, change control, and audit trails before automation is allowed to run.

For any of those, UiPath or Power Automate remains the appropriate category.

Summary

	RPA / browser automation	Hunch
Core model	Record and replay selectors	Understand fields, execute through the DOM
Field resolution	CSS/XPath/coordinates	name, id, label text, placeholder
Breakdown causes	Class renames, layout shifts, per-machine	Form removed, field renamed
Learning across sites	None — each workflow is bespoke	Network hub shares patterns across sites
Mutations	Fire and forget	Explicit visitor approval gate
Works on	Anything a robot can click, including desktop	Web pages running Hunch or owner manifests
Best for	Desktop apps, batch processing, governance	Web forms your visitors operate

If your team is evaluating Hunch against an RPA platform for web-form automation, the practical test is simple: install Hunch on one page you operate and compare what it takes to add the next form. RPA scales by recording more; Hunch scales by understanding forms.