Short answer: to choose a shared AI agent platform for your team, or an AI agent builder for Slack automation, check seven things. Decide who the agent serves, whose permissions it uses, what admins control, how pricing is counted, where the agent can run, how you leave, and whether a non-engineer can run the pilot. Runbear, which makes no-code shared AI agents for Slack and Microsoft Teams, wrote this checklist, so each step also says where Runbear fits, limits included.
A shared AI agent is one agent that a whole team uses in the same place, such as an @HR or @IT agent in a Slack channel. The checklist covers access, cost and day-to-day running. Also compare which tools each platform connects to and which AI models you can pick (on its Claude Agent SDK runtime, Runbear lets you choose Claude, GPT or Gemini for each agent). For a side-by-side comparison of tools, see Runbear's best AI agents for Slack comparison. For the in-chat questions (does it answer in the thread, read your tools, take action), see Runbear's enterprise AI chatbot checklist.
Before you compare AI agent builders
- Pick the two or three jobs the agent should do first, such as policy questions or ticket triage.
- List the tools it must read or act in, and who holds the admin account for each.
- Write down your security needs: single sign-on, data handling, and any regulated data such as health records.
- Name one person who will own the agent after launch.
The 7-step checklist for a shared Slack agent
1. Decide who the agent serves
A personal assistant works for one person. A shared agent works for a team, and everyone gets the same instructions and knowledge. Look for an agent you can name, add to a channel, and @mention there.
Runbear runs shared AI agents for teams in Slack and Microsoft Teams, where one agent serves the whole team and tools connected with per-user auth act with each person's own account. Anyone who can reach the agent in a channel where it is deployed can use it there. Dashboard access settings don't change who can talk to it in a channel. The Team plan also includes personal agents.
2. Check whose permissions the agent uses
When ten people share one agent, the question is what it can see for each of them. The vendor buyer guides from Zapier and Dust both treat permission scope as a core criterion. Ask the same thing for every tool: does the agent act as the person asking, or as one account for everyone?
Runbear lets you set each connected tool to Per-User or Shared. With Per-User, each member links their own account, and the agent acts with that member's permissions, so two people can get different answers to the same question. With Shared, one account serves everyone who uses the agent, and they all get that account's access. Keep Shared for tools where everyone should see the same data. Per-user auth covers tools, not knowledge sources: sources such as Notion or Google Drive sync through the account of the member who connected them, so connect them from an account whose access fits the channel.
3. Check what admins can control
Ask three questions:
- Can members sign in through your identity provider?
- Does the log record configuration changes, agent runs and tool calls, or only some of these? Ask for a sample export.
- Can an admin approve a tool before any agent uses it?
Runbear agents reach outside systems through MCP (Model Context Protocol), an open standard. When an Admin turns on Runbear's Require Tool Approval setting, newly connected MCP tools wait for approval before an agent can call them. Tool approvals are in limited availability, so Runbear support has to enable the feature for your organization first. An Admin sets up single sign-on (SSO) with your identity provider over SAML. Confirm with Runbear which plan includes SSO and per-user auth, and ask for the current audit log scope and a sample export.
If you need compliance paperwork, get the exact conditions in writing. Runbear's Business plan lists a SOC 2 Type II report on request. HIPAA support is covered in the FAQ below.
4. Understand the pricing unit
Per-seat pricing grows with headcount. Usage pricing grows with work, so the monthly bill is harder to predict. Many plans mix the two. Find out exactly what is counted (seats, active users, builders, credits or tokens) and what happens when you hit a limit.
Runbear's pricing page says it charges by usage, not by seat, and the work agents do is metered in credits. The Team plan costs $99 a month with 20,000 credits, and Business costs $399 a month with 100,000. Each plan also includes a set number of builders, the people who create and manage agents in the dashboard: 5 on Team and 10 on Business. Runbear's billing docs say it bills on two measures, credits and interactors (the distinct people who talk to your agents in a month), and the Enterprise plan lists a custom number of interactors, so ask how that count applies to the plan you pick. When credits run out, agents pause until you top up, turn on pay-as-you-go, or upgrade.
5. Check where the agent can run
Check which channels each plan includes, and whether the agent's runtime (the engine it runs on, chosen when you create it) narrows that list.
On Runbear, Team covers Slack, Microsoft Teams and Discord, and Business adds HubSpot and Zendesk. The runtime can narrow that list. Agents on the Claude Agent SDK runtime deploy to Slack and Microsoft Teams only, and the recommended Standard option usually creates that runtime. For a Discord, HubSpot or Zendesk agent, pick another runtime, such as Classic, when you create the agent.
6. Check how you leave
Exit risk covers the contract and the setup work your team puts in. Ask about contract length, notice period, and whether agent instructions and data can come with you.
You can cancel a Runbear plan anytime, with no annual lock-in, and choose monthly or yearly billing (pricing). If moving your agent setup elsewhere matters to you, ask Runbear about it before you sign.
7. Run a pilot without engineers
The cleanest test is one agent, one channel and one job for two weeks. Check that an ops or HR lead can set it up alone, and that answers name the document they came from.
Runbear is a no-code AI agent builder for Slack and Teams: a team lead describes the job in plain English, connects tools from a catalog of 2,000+ integrations, and deploys the agent to a channel, with no engineering ticket. Answers drawn from your documents cite the document behind them. When the documents don't cover a question, the agent is instructed to say so instead of inventing specifics.
When another kind of tool fits better
- Mostly fixed, rule-based workflows across many apps, with little chat: a workflow automation platform may fit better. Zapier, for example, lists 9,000+ app integrations and human-in-the-loop approvals.
- Each person only needs a private assistant: a personal AI assistant subscription may be enough, without a shared agent.
- Search across every system in a large company: an enterprise search product built for that job.
Common mistakes
- Judging only the demo. Ask what the agent does when your documents don't have the answer.
- Reading "usage-based" as "no limits." Ask for every measure the vendor counts. On Runbear, that means credits, builders, and the interactors described in step 4.
- Connecting every tool with one admin account. Everyone who talks to the agent then gets that account's access.
- Launching without an owner. Name one person who updates the agent's instructions and sources.
FAQ
What should a small team with no engineers check first? Start with setup that a team lead can finish alone, a pricing unit you understand before you sign, and answers that cite their source. Then check the plan's limits against how many people will ask the agent questions.
Does the vendor train AI models on our data? Runbear does not use your data to train AI models (pricing FAQ). With any vendor, get the answer in writing.
What if we handle health data? With any vendor, ask whether it signs a business associate agreement (BAA) and which setup the agreement covers. Runbear supports HIPAA compliance only for eligible Enterprise customers with a BAA and a supported configuration, and a BAA is not included automatically at signup. Before processing protected health information (PHI), confirm the agreement, organization access, and each agent's configuration with Runbear. Review connected channels, knowledge sources, model providers and tools separately for your workflow and any agreements they need. If you bring your own model API key, you are responsible for the provider BAA.
Sources
- Runbear home page
- Runbear: shared AI agents for teams
- Runbear pricing and FAQ
- Runbear features
- Runbear docs: Tools (Per-User and Shared authorization)
- Runbear docs: Members and roles
- Runbear docs: Single sign-on
- Runbear docs: Tool approvals
- Runbear docs: Billing and credits
- Runbear docs: Runtimes
- Zapier: The best AI agents for enterprises
- Dust: Top AI agent companies
