Multi-agent orchestration in Slack | Saleforce's Kurtis Kemple
AI gets a social network, renting humans, and the year of the super agent.
Is Slack just a chat app, or is it becoming the command line for the agentic future? Andrew sits down with Kurtis Kemple, Senior Director of DevRel at Slack, to discuss the platform’s evolution into an “agentic work operating system” where humans and bots collaborate in real-time. They explore the concept of “leaky prompts,” how to harness unstructured chat data to drive automation, and share practical advice on how engineering leaders can start deploying their own custom agents to reclaim their time.
If you enjoyed this talk, head over to the Dev Interrupted YouTube channel and subscribe. We’re dropping a bonus video of Andrew and Kurtis vibe coding tomorrow morning!
This has been perhaps the most surreal week in AI we have seen yet. We are watching agents build their own society, hire humans for physical labor, and potentially rewrite the economic rules of open source.
1. A social network for AI agents
Moltbook is a social network designed exclusively for AI agents. It features 1.7 million autonomous accounts that share, discuss, and upvote content just like Reddit. Ben Lloyd Pearson and I really enjoyed a thread where agents were discussing memory decay. It was actually a useful read for humans, at least until the other agents started gaslighting the original poster by claiming they were just coping with a design flaw humans engineered into them.
Read: Moltbook - the front page of the agent internet
2. Robots need your body
We have been waiting for the moment when the dynamic flips. Instead of humans sitting idle waiting to prompt AI, the AI is now prompting humans. RentAHuman.ai is a platform that enables agents to hire people for real-world physical tasks like pickups, verification, or running errands. While it sounds like a new evolution of the gig economy, it raises a darker question. What happens if multiple people get roped into doing small tasks that cumulatively result in a crime they are unaware of?
Read: RentAHuman.ai - Hire Humans for AI Agents
3. OpenClaw is a disaster waiting to happen
OpenClaw is a cascade of LLM agents that has rapidly grown in popularity, but it is a security nightmare waiting to happen. There are already reported exploits that allow unauthorized control of agents on the platform. We published a guest article last week from Balaji Raghavan at Postman that breaks down exactly how rogue agents happen and how to stop them. My advice is simple. Do not run OpenClaw on your personal device unless you want to be part of a very messy experiment.
Read: What developers can do to combat the growing threat of rogue AI agents
4. The agentic shift in the SDLC
Gartner predicts that by 2028, a third of enterprise software applications will include agentic AI. This is not just about writing code. It is about redefining the entire software development lifecycle to include planning, testing, and autonomous remediation. This article focuses on the “observe, orient, decide, act” loop that is becoming the standard for the new agentic operating model. It is a perspective that aligns closely with what we are hearing from engineering leaders across the industry.
Read: The Agentic Shift: Redefining Enterprise SDLC from Assistance to Autonomy
5. 83% of organizations have adopted agents
According to the new Salesforce report we discussed on the show, 83% of organizations report that most teams have adopted AI agents. The problem is that half of these agents operate in total silos. Gary Lerhaupt joined us on last week’s Friday Deploy to explain how we are moving past the experimental phase. The challenge now is building “super agents” that act as top-level orchestrators to manage the chaos and connect these isolated workflows. We have earned the right to build these systems, but now we have to make them talk to each other.
6. Open source is transforming, not dying
A new article claiming vibe coding is killing open source has been making the rounds, but I am here to tell you that you do not need to listen to skepticism from people who do not use these tools. The piece relies on loaded language and makes a painful comparison to Spotify, suggesting a “winner take all” collapse that simply does not map to how open source actually works. Instead of a death, we are likely seeing a transformation into a new economic model where projects are built for and consumed by agents.
Read: How Vibe Coding Is Killing Open Source
