Robert Ta walks through Vercel's Eve agent platform, which lets you deploy the same AI agent to Slack, Discord, and Teams without manually setting up bot tokens. Eve bundles durability, sandboxing, tracing, and evaluation. Jonathan adds that the platform's hierarchical skills folder and observability tools like Raindrop set new standards for agent development.
If you're starting a new agent project today, does Vercel's Eve change your stack decision? Some informed takes. Eve bundles 6 capabilities that most teams handle, like durability, sandboxing, human-in-the-loop channels, tracing, evals. One thing I really liked here is being able to run the agent locally very easily, being able to ship it easily because they're Vercel. And then this is the thing that caught my attention: introducing the agent to your team. So once you've incubated it, once you've tested it locally, once you you want to share it with your team, you don't have to go through any of the Slack BS to deploy it on Slack, like I've done with my OpenClaw and I've tested with like a Hermes agent. You could just route the credentials through Vercel Connect and there is no bot token to copy into an environment, an env file, and you just run the command for Discord, Teams, Slack, and the same agent is there across multiple services. And then you could really control the harness. I thought that was such an innovation. Like I hate doing all the bot token BS, going to Slack's interface for it. I hate that I can't just talk to Claude Code and say, handle the Slack deployment for me. And it looks like Vercel's come out with a solve for that. I think Eve does change your stack decision if you're starting a new agent project today. I think that the people that tried OpenClaw and loved it for the portability, one agent across many services, meet me at Telegram, meet me at Discord, meet me at my texts or DMs. But now you could hyper-customize this agent and you have a lot more control over the harness versus OpenClaw's standard. Super cool stuff. Jonathan, do you have anything else to add to this given your CTO perspective? Anything on the, uh, introduce agent to your team UX functionality here that I just like gushed about?
For me, this hits in like, okay, there's some new best practices that are being published and made public and they're worth paying attention to. You know, we went to the Render Localhost Conference yesterday in San Francisco. And this is where I heard about Eve, the product manager for Render was like, they just announced this. It encapsulates the best practices I'm hearing from our customers at Render, which is a strong data point. And then also mentioned like, you know, as part of this is the like trace observability, which is something that we've been rolling our own tools for, has really evolved. And he mentioned a platform called Raindrop, and that has traceability observability. I was immediately drawn to the combination of testing out Vercel Eve with Raindrop. And one of the things that led me there was like, there's this great video where the Raindrop founder is interviewing— yeah, this is it. Who's using this? Who's he interviewing? He's interviewing a Vercel product manager who uses this to run her own agents. So I think that's a nice like peek into, you know, some of the best practices that Vercel is finding, that Render's hearing about. They're kind of converging around a common story of how to use agents best practice. And I can give you a couple other tidbits that were like, no, some known best practices of how to use these agent harnesses. One takeaway I had was like that the file system is key, that you're really storing and indexing all the information on the file system for the agent to, to work with. There's a skills folder. What a lot of these agents are moving towards is skills that have like recursive nature or, or like hierarchical nature of skills where deep domain expertise itself might not be recursive yet. It's moving that way, but That's the goal. Yeah. I think it's more like you have a domain expertise that gets encoded into like hierarchical skills folders that are highly nuanced and they do a secret to the companies that run them. And so finding ways that are like basically recursive that the skills can get updated as you go is one of the ways that people are finding success. And those are things that I've had as ambitions for us, but we haven't really gotten there yet. And I think that this is like a nice kind of like best practice for us to test and see what works for us in our cases. And can we refine and train skills that are like highly nuanced and creative to the investment that we have in training our agents.
