How to Use Gumloop to Run a Weekly Competitor Watch

The short answer: a Monday competitor brief is a good first agent for Gumloop's Pro plan, which costs $37 a month for 20,000 credits (pricing page, checked on 24 October 2026): create an agent with Slack and Google Sheets connected, run the brief once by hand, ask the agent to turn the result into a skill, make Slack posts wait for your approval, add a scheduled trigger for Monday morning, and read the credit cost of the first four runs before you add anything else.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

The job is simple and easy to drop: open each competitor's pricing and product pages, note what changed, tell the team. The agent does the opening and the reading. A person still decides what matters. For what Gumloop is and how it bills, see the Gumloop glossary entry.

Before the first run:

1. Put a ceiling on it. Gumloop's credits page says one agent can cost 2 credits for a quick question and 200 for a deep research task, and works through a chat that ends at 459 credits. For example, four Monday runs at that worked example come to 4 x 459 = 1,836 credits, under a tenth of the Pro allowance.
2. Start from the list you keep. Put the competitors you already track in a Google Sheet: name, website, the pages that matter (pricing, product, careers), and one line on why each one is on the list. The agent reads that sheet each week, so editing the watch means editing a row.
3. Decide how you will judge it. After a month, count the items from the four briefs that someone acted on: a pricing change discussed, a page updated, a deal briefed. A brief nobody acts on is a cost line.

What you need first

• A Gumloop account. New accounts start with a 14-day Pro trial that needs a card and rolls into a paid subscription unless you cancel first, per the cancellation guide. Put the cancel-by date in your calendar on day one.
• Slack, with a channel the team already reads.
• The competitor sheet from the list above.

Step 1: Create the agent

A new agent asks what it should take care of and sets itself up from your answer, per Gumloop's agents documentation. Describe the job in two sentences. The builder has four tabs (Agent, Triggers, Access and Settings), and nothing saves until you press Save.

Step 2: Connect two apps, no more

Open Connectors on the Agent tab, add Slack and Google Sheets, and sign in to each. The agents documentation says the connectors you add define what the agent can see and do, and recommends starting with the two or three it needs. Connecting an app in Settings alone does not attach it to the agent, as the connector help page points out. Web Search and Web Fetch are built-in abilities, on by default, so the competitor sites need no connector.

Step 3: Make Slack ask before it posts

This is the human check, and it goes in before the first run. Open the Slack connector's approval setting and choose Ask for writes/deletes. Gumloop's human-in-the-loop page describes the result: read-only tools run freely, while a write pauses the agent and shows an approval card with the proposed message. You can approve from the chat, the notification bell, or a Slack DM. The agent reads everything it likes and posts nothing you have not seen.

Step 4: Pick a modest model

Start on the Balanced preset in the model dropdown. Gumloop's guide to reducing credit spend recommends it for everyday work and keeps Smartest for tasks that need deep reasoning. Reading five web pages and summarising them is everyday work.

Step 5: Run the brief once, by hand

Start a task and give it the whole job in plain words. For example: read the competitor sheet; for each row, check the listed pages and news from the last seven days; report changes to pricing, packaging, hiring or messaging, with a link each; stay under 150 words; draft it for the competitors channel. Watch where it guesses, and correct it in the same task until the draft is one you would post.

Step 6: Turn the result into a skill

When the draft is right, tell the agent: "Turn this into a skill." A skill holds a process the agent loads only when it is relevant, per the agents documentation. The reduce-spend guide adds a rule of thumb: instructions of more than about 200 words that do not apply to every conversation belong in a skill, because the system prompt loads in full every time.

Step 7: Add a Monday trigger

On the Triggers tab, add a Scheduled Trigger with a name, a schedule and a prompt, as the triggers documentation lays out. Describe the schedule in plain language and Gumloop writes the cron expression. Brief the prompt as you would a person, and name the skill in it so the agent loads it.

Step 8: Approve the first four posts

Each Monday the agent drafts the brief and stops at the Slack post. Approve or reject it, and when you reject, say why in the task so the agent can fix the skill. The Tasks page lists every pending card under an Approval Required filter. After four clean weeks you can tick "Don't ask again for this tool", or keep approving: a brief the whole team reads is worth reading first.

Step 9: Read the cost of each run

The running credit total sits next to each task's name, and selecting it opens a breakdown by model, tool calls, compute and the orchestration fee, per the credits page. Check the four Monday runs against the ceiling you set. If one runs long, the reduce-spend guide's first suspects are a missing skill, a model pinned too high, and tools connected but never used.

The incentive to keep in mind

Guided setup and self-scheduling make a second and third agent easy to add, and each draws on the same monthly credits. That is the business model working as designed. Add the next agent once the first one's cost per useful brief is known.

What Gumloop's docs warn about

• Vague trigger prompts. The triggers page gives "Do the usual thing" as the example to avoid: name the tool, the time window and the output.
• Connecting everything. Every tool definition adds tokens to the prompt, and the reduce-spend guide lists it as a cause of high cost.
• Credits that expire. The monthly allowance resets each billing period and does not roll over on Pro, per the pricing page.
• Too many runs at once. Pro allows 25 concurrent agent interactions across the organisation, and requests over the limit are rejected, per the rate limits page.

For the wider method behind this kind of agent, from baselines to review, see how to set up a marketing intelligence agent.

FAQ

What happens if the credits run out mid-month?

Agents stop until you enable pay-as-you-go, which is off by default, per the pricing page FAQ. Once on, extra usage bills at $0.005 a credit, and the credits page caps Pro overage at 1,000,000 credits per billing period unless you set a lower cap (checked on 24 October 2026).

Can colleagues use the same agent?

Yes. The Access tab sets owners and who can use the agent, and on Pro teammates share credentials and one pool of credits, per the agents documentation and the pricing FAQ.

Can it watch for changes during the week instead?

Yes. App triggers fire on events such as a new email in a Gmail label or a new Slack message, and webhook triggers fire when another system calls the agent's URL, per the triggers documentation. The page warns that anyone holding a webhook URL can run the agent, so treat it as a password.