Agents That Work Inside Slack: 8 Tools Compared on One Task

The best home for an AI agent is wherever your team already argues about work, and for a lot of companies that's Slack. It's where bug reports surface, account owners get looped in, and someone eventually types "who's on this?" under a message from three hours ago.

Each of these agents that work inside Slack gets one hypothetical job: summarize yesterday's #support thread, open a ticket for the bug it mentions, and tell the customer's account owner. That's three permissions in one ask (and an account owner hoping to hear before the customer emails twice).

Of the agents that work inside Slack, Lindy is the best, because it handles a thread like a coworker: it answers under your question, keeps company habits in editable memory files, and works through 1,500+ apps or any MCP server, gating shared-thread writes on Approve.

Claude Tag is a channel-wide Claude, Slackbot comes built in, and ChatGPT workspace agents carry what you built in ChatGPT. Viktor works on its own computer, and Dust routes you to specialist agents.

Zapier Agents react to Slack events, while Salesforce Agentforce in Slack serves companies that live in Salesforce.

The first route is to be part of Slack already. Slackbot arrives with your plan and needs no install, although how much of the agent you get depends on which tier you pay for.

The second is an app your admin installs. Claude Tag, Viktor, and Dust all join this way, and after that approval screen people @mention them in a channel or open a DM.

The third is an agent you build somewhere else and then deploy into channels. ChatGPT workspace agents, Zapier Agents, and Agentforce agents all start life in their home platform, which means a little design work happens before anyone in Slack sees a reply.

On the test task: You'd @mention it under the #support thread, and the summary would come back in that same thread. If your bug tracker lives in Notion, it could log the ticket right there, because Notion is part of its shared-channel toolset.

Shared channels stay public-safe by design, so the email to the account owner, or a ticket in another tool, is something you'd hand off in a DM or your own thread, where the full toolset runs. Inside the shared thread, writes you've put behind Require approval wait for someone to click Approve.

Strengths: Anyone on the team can open its memory and correct a wrong assumption, since it's stored as plain files, and recorded calls feed a shared meeting library that teammates who missed the call can draw on.

Once it has replied, a clear follow-up keeps the thread going without another @mention, since it reads the thread to work out who you're talking to.

It also sends follow-up nudges for asks that got no reply, which is handy when the account owner goes silent, and a private summary when you get pulled into a long thread.

It also opens a browser and works by hand in any tool that lacks an integration, on top of its 1,500+ integrations and MCP support.

Price: Free costs $0 and needs no card, and its $50 in credits expire after 7 days. Team runs $49.99 monthly for every active user, who each contribute 5,000 credits to one pot the team draws from, with extra credits at $10 per 1,000 and custom pricing for Enterprise.

Unused monthly credits expire, though purchased top-ups carry forward, and a nearly empty pool makes it stop credit-using actions and message you, so the invoice never jumps on its own.

Gaps: Seat billing reaches casual users. Anyone who @mentions it in Slack takes a seat once their 7-day trial ends, and returning teammates don't get a second trial, so a channel full of occasional askers adds up.

Admins do get an email for every Slack join with a one-click remove link, and they can set per-teammate monthly credit limits.

On the test task: Someone in #support would tag @Claude with the request, and because there's one Claude per channel interacting with everyone, the rest of the support team could see it working.

The ticket and the note to the account owner depend on what your admin has connected, since admins choose which tools, data, and codebases it can use.

If the thread stalls afterward, its optional ambient mode can follow up on it. If your #support room is private, though, this plan falls apart, because Claude Tag won't report from private channels.

Strengths: Memory and identities are scoped per channel, so what it learns with sales doesn't leak into engineering's setup. It works asynchronously, schedules its own tasks, and lets each channel choose its model.

Admins can cap token spend for the whole org and for each channel, review a record of each action and who requested it, and break spend down by channel. Tagging it in a channel doesn't require a Claude account by default, and you add it with /invite @Claude.

Price: Claude Tag has no price of its own. To use it as of October 2026, your company needs Enterprise or Claude Team, which charges $25 a month for a Standard seat and $125 for Premium across teams of 2 to 150, and anything done in channels gets billed to the org as usage.

Gaps: It's a beta that Anthropic launched on June 23, 2026, open only to Team and Enterprise customers. Channel usage lands on top of seat costs, and DMs run on each person's own claude.ai account with their personal connectors.

Slackbot is the agent a Business+ workspace can turn on today without a single new contract, which counts for a lot when security reviews take a quarter.

On the test task: Tagged into #support and asked to pull yesterday's thread together, Slackbot would share the recap back with citations. Slack says it can take action across apps like Jira through MCP, so with Jira connected, the ticket is a reasonable thing to ask for.

Tracking down the account owner is a search across your Slack history and connected apps, which is the core of what it does. Slack CRM also lets it create accounts, update records, and log calls if the customer record needs a note.

Strengths: It's built in, workspace owners and admins control whether it's on, and Slack promises it shows you only what your own access allows and never trains on your data.

Asking in plain language is enough to set up a repeatable workflow, and dashboards, polls, and calculators can live pinned in a channel. Skills are shareable, and mobile users get voice input.

Price: As of October 2026, Slack charges $8.75 per user monthly for Pro and $18 for Business+, while Enterprise+ pricing comes from sales. On Free and Pro, Slackbot shows up only as a trial or preview.

Gaps: On Business+, each member gets 15 Slackbot messages a week, so the person running support triage could hit the cap well before the week is out. Enterprise plans have no fixed weekly number beyond Slack's overall usage policies, and extra usage can be bought with Salesforce Flex Credits.

The full agent also means upgrading the whole workspace to Business+, which is a bigger decision than adding one tool.

On the test task: You'd do the groundwork in ChatGPT first, building a support agent there and deploying it into #support, and Slack would list it under a handle tied to a user group. For this channel you'd pick "Only reply when @mentioned" so it stays out of the way until someone asks.

Workspace agents can send emails and update documents from Slack, so the heads-up to the account owner fits.

Whether it can file the ticket comes down to its tool access, and since ChatGPT Business hooks into GitHub, Google Workspace, Microsoft 365, and Slack, a GitHub issue is the obvious thing to try.

Strengths: Agents can keep following up in the thread, join private channels after someone adds the app, and drop scheduled results into a channel. Search only covers the Slack messages and files you can already see.

On the ChatGPT side, you get usage analytics and spend controls, no training on business data by default, ChatGPT Space for working with agents, and Dots (always-on agents) on Premium seats.

Price: Seats on ChatGPT Business run $25 per user per month for Standard and $125 for Premium as of October 2026, with a 2-user minimum. You also need a paid Slack plan, and workspace agents don't work on personal ChatGPT accounts.

Gaps: Admins have three things to configure: a ChatGPT app, a Slack app, and a setting that lets members create user groups, with more on Enterprise Grid.

There's also OpenAI's own warning that agent responses may be inaccurate, which argues for a human glance at the ticket before anyone acts on it.

Viktor is the strongest fit for long, multi-tool jobs you'd like to hand off and stop thinking about until the result lands.

On the test task: One @Viktor mention in the thread would send the whole request off to its own cloud computer, and it would come back with finished work.

Linear, GitHub, and HubSpot all sit on the list of tools it can reach from one Slack message, which covers the ticket and the account lookup in one go.

Anything irreversible gets a check with you first, per Viktor's own description.

Strengths: It lists 9,800+ integrations, keeps team memory of decisions and preferences, runs scheduled automations, and offers Viktor Spaces. It's also available in Microsoft Teams, says it's SOC 2 compliant, and reports passing Slack's security review.

Price: Viktor's free tier is $0 as of October 2026, asks for no card, and includes up to $100 in credits with no expiry date. There's no per-seat fee on Team: a flat $100 a month buys 40,000 credits for everyone in the workspace, and leftover credits can roll forward one month.

Enterprise is custom, with per-user spend caps, volume pricing, and a DPA. You can also buy top-ups or set auto-top-ups under a monthly spend cap, and nonprofits pay 15% less.

Gaps: By Viktor's credit guide, a quick question in Slack costs 50 to 100 credits while a deep report runs 500 to 2,000, and this three-part job could land anywhere along that range. Recurring automations spend from the same pool, and Viktor itself points out that frequency matters.

On the test task: Dust wants you to pick the agent, so you'd mention @Dust in the thread with +AgentName (say, a support-triage agent your team built), and the reply would come back in the thread.

Only one agent answers per message, so the ticket and the heads-up go to whichever agent holds those tools, or you'd chain specialists in a multi-agent workflow.

Strengths: An admin switches on the Slack Bot under Admin > Integrations, and DMs to the app need no mention at all. Each channel can get a default agent that auto-replies with no @mention, if an admin sets one, and Slack Workflow Builder can kick agents off too.

Agents can draw on 20+ models, custom skills and knowledge, MCP servers, and dozens of connectors, and multi-agent workflows run on schedules and triggers. Frames turn results into interactive dashboards and apps, while Pods give groups a shared workspace.

For security, it holds SOC 2 Type II, supports Okta, Entra ID, and Jumpcloud sign-on, and offers US or EU data residency, with SCIM and audit logs on Enterprise.

Price: Dust prices in euros, excluding VAT. As of October 2026, a Pro seat costs €30 a month (€24 billed annually) and comes with 8,000 credits, a Max seat costs €150 with 40,000, the free tier includes 500 lifetime credits, and Enterprise is custom.

Gaps: Somebody has to design and maintain the agents, which is real setup work before the test task runs at all. Per-seat credits reset every month with no rollover.

Zapier Agents are the pick when a job can be planned in advance and kicked off by whatever happens in a channel.

On the test task: Zapier documents chatting with agents in its web app and Chrome extension, so you'd plan this one ahead of time.

You'd set up an agent that watches #support (Zapier lists a Slack Thread Insights Extractor template), creates the ticket in whichever of its 9,000+ apps you use, and posts the update back to Slack.

The note to the account owner becomes one more step in that workflow, which is easy to add once the agent knows where the owner lives.

Strengths: Zapier Copilot helps you build the agent, and agents can pull live data sources, browse the web, and use company knowledge.

Templates such as IT Helpdesk Slack Responder and Github PR Slack Notifier show how far Slack-triggered work can go, and security covers SOC 2 Type II, SOC 3, GDPR, and CCPA.

Price: As of October 2026, you get 400 activities a month free. Agents Pro is $50 a month with 1,500 activities, or $33.33 a month billed annually, and Enterprise is priced on request.

Gaps: It runs on events and workflows, so you won't get a teammate to chat with mid-thread. Every action, browse, and lookup counts toward the activity cap, and a busy support channel can burn through 400 activities in short order.

On the test task: If your admins have built a service agent, you'd @mention it in #support like you would any teammate, or DM it from the Agents tab under Tools > Agents.

It answers with your Slack and Salesforce permissions, so the account owner is the kind of record it can find, and the ticket depends on what that agent was built to do.

Strengths: Your own builders shape every agent around one job, such as fielding IT requests or explaining benefits, with Agent Script, Prompt Builder, and Agentforce Builder as their toolkit.

Prompt starters and past threads sit in the Agents tab, and you can drop a link to an agent's prompt into a channel, canvas, or workflow.

Admins assign who can use each agent, and Slack doesn't gate it by plan tier as long as there's an Agentforce license behind it.

Price: As of October 2026, an Agentforce User License costs $5 a user each month and requires Flex Credits, priced at $500 per 100,000.

Customer-facing conversations cost $2 apiece, Agentforce add-ons are $125 a user each month (Industries add-ons $150), and the top Max Edition begins at $550 per user monthly.

Gaps: You start with an empty Agents tab until someone in-house designs and sets up a service agent, so the test task waits on that build. The payoff also leans on having an Agentforce license and your customer data in Salesforce.

Before you compare features, decide how far you'd let any of these go on its own, because that answer cuts the list faster than pricing does.

Read and summarize. Every agent here can recap a thread, and reads carry the least risk, so this is a fine place for a brand-new agent to start.

Draft for a human. Let it write the ticket or the note to the account owner, then have a person send it. You catch mistakes without doing the writing yourself.

Write with a check. Approve buttons in the thread, or an agent that asks before anything it can't undo, give you a middle ground for tickets and record updates.

Write on its own. Save this for low-stakes, repeatable jobs like a scheduled recap posted to a channel, and only once the agent has earned it.

You don't need a rollout committee for this, just a short plan you'll stick to for five working days.

Day one: one channel, one owner. Add the agent to a single channel and name one person who owns it, answers questions about it, and decides what it's allowed to touch.

Days two to four: three tasks. Run the same three jobs every day, like a thread summary, a ticket draft, and a heads-up to a teammate, so you can compare results across days.

Day five: review. Look at what it got right, what someone had to fix, and what it cost in credits, messages, or seats, then decide whether it earns a second channel.

Can I use AI agents with Slack on the free plan?

Only a few, because Slackbot reaches Free workspaces only as a trial or preview and ChatGPT workspace agents require a paid Slack plan. The exception in this group is Agentforce, which runs on any Slack plan for companies holding a Salesforce Agentforce license.

Can an agent in Slack see private channels?

Some can once they're added, and the careful ones only show you what you already have access to. ChatGPT workspace agents work in private channels after the app is added, Claude Tag doesn't report from them, and Slackbot and Agentforce both answer within your own permissions.

Which agents in Slack can post on a schedule?

ChatGPT workspace agents, Claude Tag, Viktor, and Dust can all run work on a schedule. ChatGPT's agents can post results straight to a channel, while Viktor notes that frequent automations draw down its shared credit pool.