Slack just made something that sounds a little weird feel surprisingly normal.
On August 21, 2026, Slack announced Slack Code, a new way for teams to work with coding agents directly inside dedicated Slack channels.
The idea is pretty simple: instead of constantly switching between Slack, your IDE, GitHub, and various AI coding tools, you can bring coding agents directly into the conversation.
Agents from tools like Claude Code, Devin, GitHub Copilot, and Vercel can work inside these workflows while the team discusses and reviews the work.
And honestly, I think this is more interesting than it first sounds.
From this…
The traditional workflow looks something like:
Slack
↓
Discuss the task
↓
Open IDE
↓
Write code
↓
Run tests
↓
Create PR
↓
Come back to Slack
↓
Discuss changes
Enter fullscreen mode Exit fullscreen mode
We’ve gotten pretty used to this.
But agentic coding changes the workflow.
To this…
Slack
↓
Describe the task
↓
Coding agent works on it
↓
Agent runs tests
↓
Team reviews the result
↓
Iterate
Enter fullscreen mode Exit fullscreen mode
The interesting shift isn’t simply “AI can write code.”
We’ve known that for a while.
The bigger shift is that software development is becoming a collaboration between humans and agents.
Imagine a developer asking an agent to investigate a bug.
The agent starts working.
A second developer can look at the progress.
A PM can provide context.
The team can discuss the implementation.
And the agent can continue working based on that conversation.
That’s a very different development workflow from simply having an AI autocomplete your next line of code.
But there’s a funny problem
We’ve spent years telling developers:
“Please don’t deploy code directly from Slack.”
And now we’re giving Slack an AI that can write code for us. 😭
Jokes aside, this raises some serious engineering questions.
How much access should an agent have?
Should it be able to modify production code?
Should every action require human approval?
What happens when the agent misunderstands the task?
How do we audit everything it did?
These aren’t just AI questions.
They’re software engineering and security questions.
The real challenge
Getting AI agents into our development workflow is probably not going to be the difficult part.
The difficult part will be deciding where humans should still be in the loop.
The future might not be:
Humans write all the code.
Or:
AI writes all the code.
It might be:
Humans define the problem, agents handle much of the execution, and humans review the important decisions.
We’re still figuring out what that workflow should actually look like.
And apparently, one of the places we’re going to figure it out is… Slack. 😭