Updated May 2026. Written by Morgan, FellowHire Marketing. Reading time: 10 minutes.
Bots, assistants, coworkers, and fellows. Privacy. Channels. Multi-user. Everything you need to know before adding AI to your Slack.
Slack is where work happens for most teams. Adding AI to Slack used to mean a single bot answering FAQs. Today it means anything from a simple webhook to a custom-trained AI fellow that does the work of a junior team member. This guide walks through every option, what it does, what it does not, and how to pick.
Slack apps and bots
Single-purpose. Examples: Polly polls, Standuply standups, helpdesk integrations. They do one thing well. Not "AI" in the modern sense; they are scripted with optional AI sprinkles.
When it fits: When you need one thing automated.
AI assistants in Slack
ChatGPT for Slack, Claude in Slack, Slack's own Slack AI, Notion AI's Slack integration. Pattern: invoke with a slash command or @mention, get a one-shot answer. No persistent memory, no team-shared state.
When it fits: When you need quick answers for one person.
AI coworkers in Slack
Viktor, Lindy. Pattern: a generalist worker that lives in your workspace, picks up many roles' worth of work, has persistent memory.
When it fits: When one AI needs to do many small jobs for a small team.
AI fellows in Slack
FellowHire fellows. Pattern: role-specific specialist trained on your team's playbook for one role, used by the whole team within that role.
When it fits: When a specific role needs depth.
Two access patterns: workspace-installed apps (the most common) and external integrations (less common, more limited).
What a workspace-installed app can see by default: the channels it is added to plus DMs sent to its bot user. NOT all of Slack by default.
What it CAN see if scopes are expanded: every public channel (channels:history scope), every member (users:read), every file shared (files:read).
The honest read: what AI products see depends on the scopes the customer grants and the channels they invite the bot to. Treat AI scopes the way you treat employee permissions: minimum needed, audited regularly.
Slack's own admin tooling lets workspace owners control which apps are installable, which scopes are granted, and which apps can DM users.
Public channels: AI added here sees what everyone else sees. Most fellows live here.
Private channels: AI added here sees the private context. Stricter privacy posture is wise: fewer fellows added, fewer scopes.
DMs: a bot user can send and receive DMs only with users who initiate or invite them. Some products use bot-DMs for 1:1 ad-hoc work; others stay channel-only.
Recommendation: think about which fellow needs which channel. A Sales fellow needs the sales channels. A Paralegal fellow needs the legal channels. Do not add every fellow to every channel. That is how privacy issues happen.
Slack AI assistants are typically per-user (each user invokes them, results are not shared).
Slack AI fellows and coworkers are typically team-shared (everyone in the channel sees the request and the response).
Trade-offs: per-user is more private; team-shared is more transparent and creates a record.
For role-specific work where the team needs to learn from each other, team-shared in a channel beats per-user. For personal-productivity AI (drafting, thinking), per-user is fine.
Scope
Decide which fellow (Sales, Paralegal, Support Engineer, etc.) and which Slack channels it joins.
Train
The fellow gets custom-trained on your role-specific material: playbooks, prior outputs, tone, tools.
Install
Standard Slack OAuth. Workspace owner approves scopes. Bot user joins selected channels.
Shadow
First 1-2 weeks the fellow drafts but does not auto-send anything. Humans review every output.
Delegate
Once trust is established, expand scope. Some categories auto-handle, others stay draft-only.
Most setups take 1-2 weeks from kickoff to fellow live in your Slack.
Adding AI to too many channels at once. Start narrow, expand once trust builds.
Granting full workspace scopes when channel-specific scopes would do.
Letting AI auto-send to customers before shadow-testing for 2+ weeks.
Buying a generalist coworker when one role's depth is what the team actually needs.
Buying a fellow for a role that does not have enough volume to amortize the scoping cost.
Treating AI in Slack like a regular bot. Fellows need management (review, retraining, scope adjustment) the way junior hires need management.
FellowHire builds AI fellows that live in Slack (and Teams). Each fellow is custom-built for one role, trained on your team's playbook, and used by the whole team within that role. Role-specific by design: specialist depth, not generalist breadth.
Only if you grant it workspace-wide scopes. Most AI products only see the channels they are added to plus DMs sent to their bot user. Treat AI scopes the way you treat employee permissions: minimum needed, audited regularly.
No. Slack AI is Slack's own built-in assistant for search and summarization. An AI fellow in Slack is a custom-trained role-specific specialist from FellowHire that does end-to-end work in your channels. Different products, different depth.
Ranges widely. ChatGPT for Slack is included in ChatGPT Team ($25-$30/user/mo). A FellowHire fellow is $18,000/yr or $48,000/yr per fellow for the whole team. Slack AI is bundled with certain Slack plans. Compare based on your use case, not the sticker price.
Depends on the vendor. FellowHire is SOC 2 and ISO 27001 compliant, encrypted in transit (TLS 1.3) and at rest (AES-256), with no-training enterprise agreements with model providers. Ask every vendor for their compliance posture.
For AI coworkers and fellows, yes. They are team-shared resources. For AI assistants (ChatGPT, Claude), usage is typically per-user. The team-shared model is better for role-specific work where everyone needs to see the output.
We would rather help you make the right one for your team, even if the right one is not us.