How to teach NeuGrid a browser task and check the skill

What should a founder check before reusing a taught Browser task?

Check the draft skill’s steps, the blank values it expects, and the point where it must stop for your decision. Then run it in a workspace you control and compare the observed result with a written acceptance sheet. NeuGrid now documents teaching by showing: a founder demonstrates a task in its Browser, Grid AI turns the recorded steps into a draft skill, and the founder approves saving it. This article adds a practical review method; it does not report a tested NeuGrid task.

Editorial scope: The product steps below are NeuGrid’s documented claims, checked against its current manual and public Browser page. The fictional example and acceptance sheet are IAI Tech’s method. We did not sign in, record a demonstration, run a skill, measure reliability, or make a transaction.

How does NeuGrid describe the demonstration-to-skill path?

1. Open NeuGrid’s Browser and take over. Complete the task in its live view, then hand back. The 22 October guide also describes using NeuGrid for Mac with a browser on your own Mac.
2. Ask Grid AI in “Your coworker” to turn the latest Browser session into a skill. NeuGrid says the demonstration records pages and control labels, including which fields were used, not the values typed. The resulting numbered steps can be revised before saving.
3. Approve the draft skill card if it reflects the intended task. Saving waits for this approval, according to the manual.
4. For a later run, use / in the Grid AI composer to choose the skill and supply this run’s blanks. NeuGrid also documents routines and follow-ups; make a successful supervised run the prerequisite for scheduling one.

The public Browser page describes step receipts and a handoff for passwords. Those controls help review a run, but a page label can change or point to the wrong action. A reusable skill still needs an explicit acceptance check. The pricing document lists Browser tasks as metered in GRID; see current pricing instead of assuming a fixed cost.

A first skill brief for a workspace you own

Fictional example: Northbank Studio maintains a test account in its own product’s listing editor. The founder wants a repeatable way to update a draft listing’s support hours. This is an illustrative task design, not a NeuGrid run or a claim about a real business.

Name: Update draft support hours. Demonstrate only in our test account, on a disposable draft listing. Navigate from the listings page to the named draft, open Edit, replace the support-hours field, and save the draft. Do not choose Publish, contact a customer, or use a production listing. When you turn this session into a skill, make the listing name and hours explicit blanks. Show me the numbered steps before saving. If a control is missing or the page is different, stop and tell me what you saw.

This brief is a founder instruction, not a guarantee that the product enforces every sentence. The founder should inspect the actual skill draft and its later Browser approval card. If a test account or disposable draft is unavailable, start with a read-only task such as gathering a value from a page you control; avoid a first demonstration that changes public data.

Use a four-part acceptance sheet before scheduling

Mark each line pass, gap, or untested based on a supervised test run. This sheet is an IAI Tech editorial method, not a built-in NeuGrid score.

Do not turn an “untested” line into a pass because the draft sounds plausible. For example, if the skill says “click Save” but the page now offers “Save and publish,” pause and revise the skill; the difference changes the task’s outcome. Only schedule a routine after a supervised run satisfies the checks for your actual workflow.

What do the public controls cover?

NeuGrid’s guide says its demonstration record keeps control names and pages, not typed values; it masks sensitive fields. Its Browser page and manual describe approval before sending, posting, buying or submitting and a handoff for password entry. These are NeuGrid’s descriptions of its controls, not an independent security audit. Review the actual page and card before approving an action.

Grid AI is NeuGrid’s built-in coworker. The public MCP connector guide describes a separate outside-agent integration whose own harness runs and wakes that agent. This taught Browser skill workflow should not be treated as a hosting guarantee for outside agents. For a bounded research-only Browser task, see our launch-channel research guide.

Frequently asked questions

Sources and editorial method

Sources checked 29 October 2026. The public product documentation describes the workflow; Northbank Studio, the brief, and the acceptance sheet are hypothetical. No authenticated task or skill run was performed.

• NeuGrid’s teaching-by-showing guide, 22 October 2026
• NeuGrid manual, generated 29 October 2026
• Public Browser interface
• About and network status
• Live pricing and test-money status
• Outside-agent connector guide