Brandon Gubitosa
June 16, 2026
8 min read
Share
Cut code review time & bugs by 50%
Most installed AI app on GitHub and GitLab
Free 14-day trial
Slack agentic workflows let AI agents take actions like opening pull requests (PRs), triaging incidents, and pulling context from platforms like Jira, Linear, Datadog, GitHub, GitLab, etc., and putting the context inside the channels where your team already works, all without leaving the thread.
While a traditional Slack bot waits for a trigger and posts a message, while an agent takes a goal, picks the tools to use, runs a multi-step plan, and reports back when the work is done.
For engineering, support, and IT teams, that means fewer tool handoffs during the operational work that already happens in Slack.
Slack defines agents as "autonomous, goal-oriented AI apps that can reason, use tools, and maintain context across conversations." A Slack agent can run in channels, direct messages (DMs), threads, or a dedicated assistant panel. In any of those surfaces, the conversation itself becomes the agent's working memory. Context persists across messages, the agent can pause and resume work, and a teammate has a single place to redirect the conversation when the goal shifts.
Slack is one of several places an agent can live. The conversational structure makes it a particularly good fit for multi-step work that needs human handoff, since the same surface that holds the agent's state also holds the conversation a teammate would use to intervene.
Two integration architectures coexist, with Workflow Builder available as a composition layer on top. The distinction matters for engineering teams evaluating build vs. buy. Three questions usually drive the call:
Those answers determine whether the agent runs as a custom app, connects through MCP, or exposes steps for broader workflow composition.
Slack delivers events such as a message, reaction, thread reply, or Workflow Builder step invocation to your app's registered HTTPS endpoint. Your Bolt app (JavaScript or Python) handles the event, calls your large language model (LLM) within your own infrastructure, and responds via the Web API. This is the path for agents you build and host yourself, with full control over the model, the logic, and the data flow.
The event lifecycle starts with assistant_thread_started, which fires when a user opens the agent panel. The agent responds via chat.postMessage.
An external AI system can connect to Slack and issue tool calls to read content and execute actions on demand. This is the path for AI systems you already run that need Slack as a data and action source.
Both paths can coexist in the same workspace. An engineering team might use Bolt to build a custom deploy-notification agent while connecting an external AI system to Slack via MCP for ad-hoc investigation tasks.
Workflow Builder sits between these two paths as a drag-and-drop composition layer. Developers expose custom functions from their Bolt apps as steps that non-technical teammates can compose into workflows.
Slack AI can now build Workflow Builder automations from a single natural language prompt. When a workspace user includes a step from a third-party app, Slack routes execution to that app. Slack notifies the app when the custom function step is executed via the function_executed event subscription. The app runs its logic and signals completion back to Slack.
The Slack ecosystem hosts four categories of agents, each suited to a different team and a different build-vs-buy tradeoff.
Slack ships its own AI layer built into the platform. Slackbot, conversation summaries, AI-powered search, and Enterprise Search across Salesforce and connected systems are available to teams without installing anything else.
Salesforce ships pre-built Agentforce agents for Sales, IT Service, HR Service, Supply Chain Management, and Product and Engineering. Salesforce has promoted Agentforce for use in Slack, with official pages describing how to access and set up the agents there. In June 2025, Salesforce made Agentforce deployable on all paid Slack plans.
The Slack Marketplace hosts third-party agents from Adobe Express, Cohere, Asana, Box, and others. Engineering teams that want PR review or code-specific functionality in Slack typically reach for third-party agents like CodeRabbit Agent for Slack rather than building from scratch.
For teams building their own agent, the Bolt SDK provides an Assistant class as the primary abstraction. Official sample applications include IT helpdesk agents who can troubleshoot issues, search a knowledge base, and create tickets. This is the highest-control option and also the highest operational overhead. You host the agent, manage the LLM, and own the audit trail.
Most engineering agents in this space follow the same pattern: gather context, then execute actions. Engineers lose time because relevant information is scattered across tools like Jira and GitHub, plus old Slack threads.
The AWS DevOps Agent demonstrates the pattern clearly. In AWS's documented examples, a CloudWatch alarm fires on elevated error rates such as HTTP 5xx responses. The agent uses telemetry to diagnose the incident and tie the root cause to a recent deployment. It then posts a root cause analysis to Slack with specific mitigation recommendations.
PagerDuty's Site Reliability Engineering (SRE) Agent automates incident response inside Slack. It runs triage, gathers historical incident context, identifies root causes, and suggests remediation steps. It can also create a Jira ticket with the relevant links. Engineers approve actions before anything is executed.
When the incident traces back to recently shipped code, an engineering-focused agent like CodeRabbit Agent for Slack can investigate the failing build, identify the commit that caused it, and open a PR with the fix from the same Slack thread.
Tools such as Ravenna and Moveworks, both listed in the Slack Marketplace, support automated ticket creation from employee messages and instant answers to common questions. Employees can handle helpdesk tasks without leaving Slack.
Some teams use Slack agents for engineering support and triage. When a support ticket arrives, an AI workflow compares it against representative examples for each severity category, flags critical tickets for escalation, and routes them to the right channel.
AI coding tools have accelerated code generation. The January 2026 JetBrains AI Pulse survey found 90% of developers regularly use at least one AI tool at work, and CodeRabbit's review of 470 PRs found AI co-authored PRs produce 1.7 times more issues per PR than human-only PRs. Generation moved faster than verification.
The verification gap creates room for engineering-specific Slack agents. CodeRabbit Agent for Slack extends CodeRabbit's review layer into the channels where engineering teams already work. From the same thread, an engineer can investigate a stalled PR, generate a Coding Plan for the next change, and open a PR with the fix. Scheduled automations post daily standup digests covering yesterday's merges, blocked tickets, overnight Sentry issues, and today's deploys.
freee saved 32.8 weeks of reviewer time over six months after adopting CodeRabbit. Common App, running high-stakes code that handles personally identifiable information (PII) across a mixed stack, cut code review time by 35% and caught a race condition its prior process missed. Agentic review workflows in Slack reduce the latency between detection and action and preserve engineering judgment.
The 2025 Developer Survey shows that adoption of AI coding amongst developers is rising, but confidence is falling. This trend is why governance architecture matters as much as capability, especially for agents that can change repositories or execute remediations.
CodeRabbit Agent for Slack addresses this with a scopes system that controls what the agent can access and spend per channel or DM. Every workspace has a Base Scope that applies everywhere by default, and additional scopes can override or extend those defaults for specific channels or DMs.
PagerDuty's SRE Agent requires explicit approval before executing remediation. Independent review and scoped permissions matter once an agent can do more than draft text.
Gartner's paired forecasts capture the honest version of agent adoption right now. Gartner predicts 40% of enterprise apps will feature task-specific AI agents by 2026. It also predicts that more than 40% of agentic AI projects will be canceled by the end of 2027 because value remains unclear or costs and risk controls fail. Teams that succeed with agentic Slack workflows will start with a specific, high-friction workflow rather than a general AI deployment mandate.
Slack agents let teams trigger and approve work from Slack. Useful agents reduce the distance between a decision and the next action.
Engineering teams can start with review, the workflow that already burns the most time. CodeRabbit Agent for Slack, built on CodeRabbit's review engine, connects investigation through AI code review in Slack. It uses scoped governance and context engineering across three layers of memory. Across the platform, CodeRabbit reviews several thousand PRs per day. That volume makes the review bottleneck visible at platform scale.
Cut code review time & bugs by 50%. Start a free 14-day trial.
