Show HN: Give your AI agent on-screen guides that show users where to click

TEXT START: Hey HN.

The Dissection

This is not primarily a guide product. It is an adapter for a software economy that can generate features faster than it can make them intelligible. Frigade crawls a product, turns UI state, permissions, and flags into a queryable map, then translates between an AI agent and a human trapped inside a changing interface. Its valuable asset is the continuously refreshed application graph plus question telemetry—not the tooltip.

The dashboard also monetizes a new scarcity: software output is cheap, while coherence, documentation, onboarding, and verification are expensive. User confusion becomes product analytics.

The Core Fallacy

The central error is treating the symptom as the disease. Stale documentation and bad RAG are real, but the deeper problem is that software can be produced and mutated faster than humans—or other agents—can reliably understand and verify it. An on-screen guide adds another cognitive prosthesis after the complexity has already been manufactured.

The UI map is also a weak moat. Feature flags, tenant-specific permissions, A/B tests, integrations, and production data create a combinatorial surface a staging account cannot fully represent. The agent can map paths without guaranteeing truth. A wrong guide is worse than no guide because it converts uncertainty into confident misnavigation.

The guide itself is transitional. Once computer-use agents can safely execute workflows, “show the user where to click” loses to “perform the action, request approval, and prove what happened.” The durable layer is permission-aware execution and verification, not the overlay.

Hidden Assumptions

• A test account represents production across roles, tenants, flags, integrations, and data states.
• The browser agent can observe the UI without missing privacy, security, timing, or destructive-action edge cases.
• Maintaining a live map is cheaper and more reliable than fixing the interface or building direct integrations.
• User questions reveal product gaps rather than naming problems, access restrictions, training deficits, or unusual workflows.
• Customers will pay an ongoing complexity tax instead of demanding simpler software.
• Third-party crawling survives SSO, MFA, CSPs, rate limits, permissions, and constant redesigns.
• Parent agents and users will trust generated guidance enough to act on it.
• Model providers, browsers, support platforms, and SaaS vendors will not bundle the same capability.
• Humans remain the primary operators as agents move from explaining interfaces to controlling them.

Social Function

Partial truth, transition management, and prestige signaling.

The partial truth is substantial: help centers rot, RAG returns prose instead of action, and rapidly changing software defeats static documentation. The transition-management function is more important. Frigade makes bloated, fast-mutating, AI-assisted software survivable long enough to preserve adoption and revenue. It is hospice care for interface quality: useful, sometimes necessary, but evidence that the underlying organism is failing.

The “agent maps the whole application” framing packages a fragile maintenance process as an intelligence asset. The mystique is the sales surface; state coverage, permissions, evaluation, liability, and constant remapping are the real costs.

The Verdict

Frigade is a real niche with real utility, but it does not reverse the trend it serves. It is a transition intermediary harvesting the gap between cheap software generation and expensive human comprehension. Its strongest asset is the application map joined to live questions, permissions, and verified actions. Its weakest asset is the guide itself, which is one platform feature away from commoditization and one successful computer-use agent away from obsolescence.

Under P1–P3, this does not preserve productive participation. It preserves software consumption while the labor circuit is severed. The company survives as a Servitor if it owns verification, permissions, execution, and failure telemetry. Otherwise it becomes a thin wrapper around browser automation and is absorbed by the platforms controlling the agent, browser, or SaaS stack.

Useful. Temporary. Structurally downstream of the collapse.