A comparison of several products often starts with independent research. Each product has its own announcements, pricing pages, documentation, and availability details. Bringing those findings into one useful report requires a shared method and a final review.
AutoClaw Agent Cluster organizes complex work around that process. It reviews the request, prepares a plan, assigns suitable work to specialized agents, and brings their findings together. Outputs containing conclusions, code, or actionable recommendations go through review before delivery.
This guide uses an illustrative product comparison to explain how to brief the cluster, follow its work, and review the final report.
What is AutoClaw Agent Cluster?
Agent Cluster is a mode for handling work with multiple stages or dimensions. It adds an explicit planning and review process to AutoClaw’s ability to research, use tools, and produce deliverables.
When parts of the task are independent, specialized agents can work on them in parallel. The main agent combines the findings into a coherent output. The documented workflow also calls for checking key facts online and reviewing outputs that include conclusions or executable work.
For a product comparison, research on separate products is a natural division. Each researcher can gather evidence while using the same categories. The comparison and recommendation depend on those findings, so they come after the relevant evidence is available.
The number of roles varies with the task. A small assignment may need only a few roles. More roles are useful only when the work benefits from that division.
Give every research stream the same rules
Before enabling the cluster, define the question the report should help answer. A team choosing a tool needs different evidence from a team tracking recent releases. Specify the products, the reporting period, the sources, and the output format.
Then choose consistent comparison fields. Without them, one section may focus on pricing while another describes features, leaving the reader unable to compare the products directly.
| Field | What the research should record |
|---|---|
| Capability | What the official source says the product can do |
| Availability | Any stated plan, platform, region, or rollout restrictions |
| Pricing | The plan and price shown, with the observation date |
| Evidence | The source link and relevant publication or access date |
| Uncertainty | Missing information, inaccessible sources, or conflicting statements |
These fields give independent research streams a common structure. They also make gaps visible. If a product’s documentation does not state a restriction, the report should explain what could be verified instead of filling the gap with a guess.
Start the task in Agent Cluster
Select Agent Cluster in the conversation input area, attach the relevant product list or background materials, and describe the task. Include the deliverable and the criteria you want used throughout the comparison.
Compare the five AI products in my attached list for a small marketing team. Use official product pages, documentation, pricing pages, and announcements. Research each product using the same categories: supported workflows, availability, pricing, and relevant limitations. Record source links and dates. Divide independent research where useful, then combine it into an editable report with a comparison table. Review conflicting information and unsupported conclusions before delivery. Identify anything you could not verify.
This prompt defines a shared research method and a finished output. It leaves the cluster free to choose an appropriate division of work while keeping the final report consistent.
For a release-tracking report, add a clear time window and a rule for what counts as a new development. For a purchasing decision, include the team’s actual requirements and budget. Those details make the research relevant to the decision.
Follow the plan and role progress
Agent Cluster displays the task plan and progress so you can see the stages of the work and the roles involved. Use that visibility to check whether the task is moving in a useful direction.
If the plan includes a broad market survey but you only need a comparison of five named products, clarify the scope early. If the task depends on a specific document, provide it before the agent builds conclusions around incomplete information.
Parallelism works best when each role can proceed with the inputs already available. Research can be divided by product, topic, or source set. Final synthesis needs to use the collected evidence, and review needs a draft to inspect. The workflow should respect those dependencies.
Review the combined result
The final report should help a reader compare the findings without reconstructing every research stream. Look for consistent categories, usable source links, clear dates, and explanations of material uncertainty.
Pay particular attention to statements about change. A pricing page can show the current price without establishing when it changed. An announcement can describe a feature while access remains limited to certain users. Those distinctions affect the conclusions the report can support.
For a recommendation, ask whether the evidence connects to the team’s requirements. A product having more features does not by itself show that it suits the team’s workflow. A useful recommendation explains which requirements it meets and which tradeoffs remain.
If a gap appears, request a focused revision. For example: “Check whether the collaboration feature is included in the listed plan, add the source, and revise the recommendation if necessary.” That gives the next research round a concrete purpose.
When is a cluster workflow useful?
Agent Cluster fits broad research, document synthesis, financial analysis, and substantial writing tasks where parts of the work can be divided and the result benefits from review. A literature review might split evidence collection by topic; a report might combine analysis with chart preparation.
For a quick lookup or a short wording change, ordinary conversation is usually sufficient. Cluster Mode introduces a fuller process, so it is most useful when the assignment warrants planning, coordination, and synthesis.
Can I choose an exact number of agents? The documented workflow selects roles according to the task’s complexity. Give it a clear scope and useful divisions of work; role count alone is not a measure of the final report’s quality.
Does parallel research remove the need to check sources? Source review remains part of assessing the deliverable. Inspect the evidence behind material claims and any reported access gaps.
For more examples of supported task types, read the AutoClaw Cluster Mode overview. To try the workflow, download AutoClaw and start with a research question, a bounded source set, and one consistent comparison structure.
