Research agents are useful when they can search broadly, read the primary material, reconcile disagreements, and tell you what remains unverified. Strawberry is designed to take a research question through the browser and return findings that a teammate can inspect rather than a polished guess.
01
Start with the decision that the research must support.
“Research the market” is too broad to check.
“Identify twenty European payroll platforms with public pricing, a self-serve trial, and an enterprise offer” gives the agent a scope, fields, and a concrete stop point.
02
Separate discovery from proof before you synthesize.
Search results can surface candidates, but product pages, public filings, documentation, and direct statements are where decisive claims should be verified. A good agent records the source beside the finding and flags a conflict instead of averaging it away.
03
Ask for the shape of the output before the search begins.
A board memo, competitor table, vendor short-list, and customer brief need different fields and different levels of certainty. Naming that output in advance prevents a wide web sweep from becoming a document nobody can use.
04
Use repeatable briefs when the question comes back.
Skills are useful for research that follows the same evidence rules every time, such as weekly competitor monitoring or a recurring account brief. Routines can trigger the work at a chosen cadence while the underlying acceptance criteria stay visible.
The best choice depends on whether the agent can find primary material, preserve sources, and expose uncertainty for the question you need answered.
It can when the research workflow requires sources alongside the findings. Citations should point to the page that supports the claim, not a search-results preview.
Set a source hierarchy, require evidence for material claims, and ask the agent to flag missing proof or conflicting sources instead of filling gaps.
Yes. A skill can hold the research question, source rules, output format, and QA checks so the next run begins with the same method.
Yes. A routine can repeat a defined research check and return new or changed evidence, subject to the scope and access you set.