Slack AI Agent for Business: Buy or Build One?

A Slack AI agent is a piece of software that sits in your team's channels like a colleague: it watches for requests, does the routine part of the work in your other systems, and messages a person when it needs approval or a decision. You can now rent one off the shelf (Perplexity launched "Teammates" on 5 November 2026) or have one built around your own tools and rules. Rent one if the work belongs to one person and lives in public web tools. Build one if the agent serves a whole team, touches customer or financial data, or needs to act inside your own systems. A focused custom build typically costs $8,000 to $20,000 and goes live in four to six weeks.

This post is part of our guide to rolling out agents, which starts at wizeb.com/blog/how-to-implement-ai-agents-in-business.

The News: AI Teammates Arrive in Slack

On 5 November 2026 Perplexity released Teammates for subscribers on its Max plan, on the web only for now. According to AlphaSignal's write-up (alphasignal.ai/news/perplexity-s-teammates-gives-ai-agents-persistent-roles-and-memory), each teammate gets a role, standing instructions, tools and escalation rules. It keeps one running thread of what it has done, and it messages coworkers in Slack when it needs approval, clarification or a decision. Each one runs on its own isolated computer with a separate browser and cookie store, so a marketing teammate can stay signed in to an ad account while an engineering teammate works in GitHub.

Pricing follows the Max plan, which AlphaSignal lists at $200 a month with 10,000 credits. Teammates draw on those credits, and monitoring, browser work and parallel tasks use more of them. ShiftDelete's coverage (shiftdelete.net/en/perplexity-teammates-ai-agents) says Enterprise Pro and Enterprise Max accounts are not included by default, and that each teammate inherits the user's own projects, plugins, memory and permissions. Neither source gives customer or usage numbers yet.

The product matters less than what it signals. The chat window you type into is turning into an agent that lives in the place your team already works, picks up requests without being asked, and only interrupts a person when it has to. Owners are going to be asked about this by their staff within months, so it is worth deciding now what an agent in Slack should and should not be allowed to do.

The Problem: Slack Is Where Requests Go to Wait

In most growing businesses, Slack has quietly become the request queue. Sales asks ops for a delivery date. Support asks finance whether a refund is approved. A manager asks for last week's numbers. Someone copies data from three tools into a reply, and the same question comes up again on Thursday. None of it shows up in a ticket system, so nobody sees how many hours it eats.

That is exactly the kind of work an agent is good at: frequent, rule-based, spread across systems, and mostly answered from data you already have. The question is not whether to put an agent in Slack. It is whose agent it is and what it can touch.

What a Slack AI Agent Can Do for a Team

1. 1Triage a request channel. It reads new posts in #ops-requests or #help-finance, tags them by type, answers the ones it can from your data, and routes the rest to the right person with the context already gathered.
2. 2Answer status questions. "Has the Acme order shipped?" "Is invoice 4417 paid?" It checks the ERP, CRM or accounting system and replies in the thread, with a link to the record.
3. 3Prepare routine reports. Every Monday it pulls pipeline, cash and support numbers into a short summary for the leadership channel, and flags anything that moved more than you said it should.
4. 4Draft and route approvals. A discount request, a refund over the limit or a new supplier goes to the approver as a Slack message with approve and reject buttons, and the agent finishes the job once someone decides.
5. 5Watch for problems. It notices a spike in failed payments, a stuck order queue or a missed SLA and posts it to the right channel before a customer complains.
6. 6Hand off to other agents. A support agent passes a confirmed bug to an engineering channel with steps to reproduce, or a sales agent hands a signed deal to onboarding.

Off-the-Shelf Teammate or Custom Slack Agent?

Rented teammates such as Perplexity's are a good way to learn what an always-on agent feels like, and for one person's research, monitoring or admin they may be all you need. The limits show up as soon as the agent works for the business rather than for one person.

• Whose permissions it uses. A teammate that inherits one user's accounts can do whatever that user can do. A team agent should run under its own service account with only the access its job needs, so leaving staff do not take it with them and a mistake has a ceiling.
• Which systems it can reach. Browser-driven agents are good at web apps. Your ERP, a private database or an on-premise system usually needs a proper API connection, built and tested once.
• Where your data goes. A UK or EU business handling customer data needs to know where it is processed and logged, and whether the vendor trains on it. A custom build lets you choose the model provider and region.
• What it costs at volume. Credit-based pricing is easy to start and hard to forecast once an agent watches several channels all day. A custom agent's running cost is model usage plus hosting, which you can measure per request.
• Who can audit it. You want a log of every action the agent took, in which system, on whose request, and who approved it. That is a requirement for most finance and HR work.

Our full decision framework is at wizeb.com/blog/build-vs-buy-ai-agents-2026-decision-framework. The short version: rent when the job is personal and low-risk, build when the agent touches shared data, money or customers.

How to Set Up a Slack AI Agent in Five Steps

1. 1Pick one channel and one job. Look at a month of messages in your busiest request channel and count the repeat question types. Choose the most frequent one that is answered from data, not judgment.
2. 2Write the rules. What the agent may answer on its own, what needs approval, who approves, and what it must never do. One page is enough.
3. 3Give it its own identity. Install it as a Slack app with a named bot user, connect it to each system with a service account, and grant read-only access first. Our permissions checklist is at wizeb.com/blog/ai-agent-permissions-least-privilege-2026.
4. 4Test it on history. Replay 100 to 200 past requests and score its answers before anyone sees it live. Agree a pass mark up front, for example correct on 90% with no wrong actions.
5. 5Run it in suggest mode. For the first two weeks it drafts replies in the thread for a person to confirm. Switch on direct replies for the simplest request types once the drafts hold up, then add actions one at a time.

What It Costs

• Rented teammate: from about $200 a month per user on Perplexity's Max plan, plus the time to configure roles and connections. Usage limits apply through credits.
• Custom agent, one channel and one or two systems (for example, order-status answers from your ERP in #sales-ops): about $8,000 to $20,000 to build, four to six weeks.
• Custom agent across several channels with approvals and actions: about $20,000 to $45,000, eight to twelve weeks.
• Running costs for a custom build: model usage, hosting and monitoring usually come to $150 to $800 a month for a team of 20 to 100 people.

These are WizeB's working ranges for growing businesses in the US, UK and Europe, not a price list. The full breakdown is at wizeb.com/blog/ai-agent-cost-guide.

Example: An Ops Channel at a 45-Person Distributor

This is an illustrative example, but the pattern is common. A US wholesale distributor with 45 staff runs everything through Slack. Its #ops-requests channel gets about 600 messages a month, and two operations coordinators spend most mornings answering them. Counting a month of messages shows that 55% are order-status and stock-availability questions, all answered from the ERP and the shipping portal.

Two coordinators try a rented teammate first and like it for their own research, but it works under their personal logins and cannot reach the ERP. The company has a custom agent built instead: a Slack app with its own bot user, read-only access to the ERP and shipping data, and a rule that anything involving a price change or a credit goes to the ops lead for approval. After two weeks of drafts, it answers stock and status questions directly. The build costs about $14,000 and runs at roughly $300 a month. The coordinators now spend their mornings on the exceptions, and the sales team gets answers in minutes instead of by lunch.

How WizeB Builds Slack Agents

WizeB builds custom AI agents, workflow automations and voice agents for growing businesses in the US, UK and Europe. For Slack, we start from your busiest request channel, pick the one job that saves the most hours at the lowest risk, connect the agent to your systems with its own scoped accounts, add approval steps where money or customers are involved, and test it against your own message history before launch. You own the prompts, workflows, test sets and accounts. Details are at wizeb.com/services/ai-agents.