Work continues in the browser
Improving a service includes reading pages, trying inputs, and checking web behavior after implementation. Logged-in services also require account access.
If a person takes over every browser task, the mission breaks between implementation and checking results. Delegating those tasks requires a clear boundary for login credentials.
Open, read, fill, and test pages through Web Tabs
In the October 14, 2026 release, Web Tabs let agents open, read, fill, and test pages inside Marblo. Agents can work on delegated logged-in services while cookies and passwords remain unexposed to them.
Delegation covers page tasks in logged-in services, from opening and reading content to entering information and testing results.
Separate task delegation from sharing credentials
Delegating browser work can reduce manual relays: copying page content for an agent, entering information on its behalf, or performing checks. Web results can then be examined within the same task context.
Cookies and passwords remain unexposed to the agent. These credentials grant account access; copying them into a task description shares that access material alongside page content. Web Tabs keep credentials unexposed when you delegate work in a logged-in service.
Teams define which page information an agent should read and what it should enter. Credential handling leaves those choices to the assigned task’s purpose and scope.
When to use it
Delegate these page tasks:
- Checking a web page after implementation: delegate reading the page, trying inputs, and testing the expected result.
- Continuing an input task in a web service: identify the steps where a person currently relays page content or enters information for an agent.
- Working in a logged-in service: delegate the service task without putting cookies or passwords in the task description.
Specify what to check and the expected result to tie page work to the goal.
Connect it to the closed loop and independent verification
The closed loop connects assignment, verification, review, and merging within a mission. For missions involving web pages, define code changes and expected browser results together.
Risk-based independent testers use a separate AI tester to verify important changes, including counterexamples. Define page tasks alongside independent verification to specify the work and the results to check.
In this series
- Marblo Build Notes 1 — The Closed Loop Release
- Marblo Build Notes 2 — Completing a Mission with the Closed Loop
- Marblo Build Notes 3 — Agent Mosaic
- Marblo Build Notes 4 — Orchebrain
- Marblo Build Notes 5 — Web Tabs
- Marblo Build Notes 6 — Merge Train
- Marblo Build Notes 7 — Multiple AI Accounts
- Marblo Build Notes 8 — Risk-Based Independent Testers
- Marblo Build Notes 9 — Telegram Conversations with the Orchestrator