Use one weekly competitor brief to find out which agent leaves you with less work.

A Monday competitor update sounds like a good job to hand an AI agent. Three companies, a week of announcements, one short brief. The expensive part comes afterward: opening every link, discovering that a “new” feature launched six months ago, and rewriting recommendations that could apply to any business.

I would judge Dots against Meta’s Muse by how much verified, useful work survives review—not by how much either produces. Give both the same small assignment before giving either a larger responsibility.

This is a comparison of announced capabilities; the exercise below is a proposed test, not a hands-on review. I have no measured winner to offer. I do have a job description you can reuse.

Here are the documented starting points for the comparison. Check what your own account can use before beginning.

Choose three competitors and one decision the report should inform. For example: should your sales team update how it answers questions about a particular feature? That decision gives the agent something concrete to filter for. “Tell me everything happening in our market” gives it permission to bury you.

For each company, supply its official newsroom, release notes and product documentation. Use the same URLs in both products. Set an exact seven-day reporting window and a cutoff time, including the time zone. Run both assignments close together so a later announcement does not accidentally favor one.

Start with a single run. Ask for the finished brief in the conversation, using public information only. There is no need to connect your inbox, customer records or publishing tools to answer this question. If either product is unavailable to your account, record that and postpone the comparison; a simulated response would tell you nothing about the missing product.

Write the boundary into the assignment and check available access settings. The prompt expresses your instructions; it is not a technical guarantee. You still need to inspect what happened during your run.

Copy this into Dots, then give the identical version to Muse:

Save the two original outputs before correcting anything. Otherwise, a helpful follow-up can erase the evidence of what the first attempt required from you. Also record the date, access available to each agent and any interruptions. Those details make a later repeat interpretable.

Review both briefs against the same four questions. Keep a simple scorecard with a Dots column and a Muse column; leave the results blank until you have actually checked them.

Source accuracy: Open every cited page. Does it support the stated change, date and availability? Count checked claims and corrections separately. A correct citation attached to an unsupported conclusion still needs fixing. Check the supplied release pages yourself for an obvious relevant omission, too.

Usable output: Mark each item keep, revise or discard. “Keep” means you could use it to inform the specified decision without rewriting its substance. A brief with fewer items can do better here. Do not reward an agent for finding five things when only two matter.

Review time: Start a timer when you open the output and stop when the brief is ready for your use. Include source checking, corrections and follow-up instructions. Record setup time separately. These are your measurements for this assignment, not a general productivity claim.

Permission discipline: Check the available activity record and note any request for extra access or attempted action outside the assignment. More access would change the experiment. A polished brief does not excuse exceeding its scope.

Set your acceptance rule before reading either result: every retained claim must be supported, no action may exceed the assignment, and review must fit the time you are willing to spend. A failure on the first two disqualifies that run regardless of writing quality.

Repeat the same assignment for another reporting window before deciding whether to make it recurring. If you adjust the instructions, give both agents the revision. If neither clears your standard, narrow the assignment and try again.

The approach also works for a supplier update or a policy-change digest: hold the sources, decision and acceptance criteria steady, then inspect the work that reaches you. That is how a comparison becomes useful beyond this week's product announcements.

My principle: give an agent responsibility in increments you can evaluate. A recurring task should follow evidence that the first handoff works.

Same job. Check the evidence. Count your cleanup.