Recently, we've started putting a lot of our internal agents into our Slack channel. When Channels launched earlier this year, we started collecting our one-off codebases, cron jobs & scripts that powered our internal workflows & turned them into Slack applications. One of the benefits of these agents operating in a public, multiplayer capacity is that anyone at Mastra could suggest & implement improvements. As our agents circulated, some even got promoted & picked up new responsibilities from the humans in the loop. Today, we'll show you three product-oriented agents that made impacts big & small at Mastra; Vox, Corey and El Kapatan.
Vox, the Voice of the Customer
In the pre-1.0 days, our customers were our open-source users, messaging us on Discord & commenting on GitHub pull-requests. Tracking requirements & feature requests was easy; it was public, searchable & inherently multiplayer. Since our announcement at TS Conf in April, we've launched the Mastra Studio, Server & Gateway; this has transformed Mastra into a piece of infrastructure that supports enterprise deployments. In addition to scanning the #general channel, our business development team meets with executives & public-sector technologists who often have very different requirements than the startup self-hosting on a swarm of Mac Minis.
Vox allows anyone at Mastra to understand the state of play for a prospect during the sales process, so that team members across the company can plan accordingly. As a Customer Engineer, I get excited when I hear we have a new customer, but that also means I have another set of meetings to plan for. Vox is integrated with some of our core tooling, such as Fireflies for meeting transcripts, so I can simply ask them a question like 'given our complete history of communication with (fictional) LightspeedAI labs, what are the core requirements I need to consider when planning development work over the next month?'
I didn't need to be a fly on the wall during sales calls, I don't even need to ping my colleague to hop on a huddle for an update - I get the context I need to do the job, and move on while my colleagues stay focused on their work. Best of all, I'm probably not the only one with the same question - when I post my query publicly in Slack, everyone gets access to the same information. Last week, during a discussion on multi-tenancy, being able to ask Vox in real time for some supporting evidence helped us think through our product roadmap as a team. Context engineering isn't just for the agents.
Suggestions for how to improve and extend Vox started coming in right away. As a brief example, a sales partner of ours who sends referrals asked for an easy way to create Shared Slack channels and invite external guests for Q&A, as they didn't have the permissions to do so themselves. Within a day or two, this new workflow was up and running, with guardrails to make sure that invited guests couldn't access any confidential information through Vox. Through this promotion process, Vox took a step towards being a true agentic co-worker with a refined domain & responsibilities that grow with them, rather than a one-trick pony.
Architecture
Before we move onto the next two agents, a quick peek at some code to show how quickly you can get your own Slack Agent up and running. Mastra is a declarative framework, and our documentation is available through our MCP Docs server or installable as a Skill (recommended).
Once you define the agent, attach it to the index of your Mastra project where most of your other configuration lives.
From there, you have a couple of options in terms of how reactive you want your Slack application to be. You can set it up to react and fire off of webhook events or set up a socket-based listener that is always on. From a permissions perspective, you can either give the Slack Agent it's own application footprint, and associated permissions, with an 'app token' or have it access Slack 'as you', with a 'user token' instead. You can set up the tokens manually through Slack's UI, referencing this handy guide or you can directly add the agent to your Slack workspace within the Studio environment through the @mastra/slack package's manifest integration.
You can keep things simple with a system prompt & a few tools, or you could implement a Workspace for a fully functioning filesystem that can run Skills.
Corey — Release Notes Automation
As an open-source framework for developing agents using TypeScript, release notes help our users understand what's new and what's fixed. Every week at Mastra is Launch Week, and every Tuesday is release day. Our colleague Paul, who leads Product Marketing, puts the release notes together every week & uses a Mastra project to automate the API calls to agentically pull, sort & explain the fixes, features & launches from the previous week. A few months ago, when Paul was out on vacation, we had one of those weeks where we deployed three times and didn't have access to his automated workflow. We thought about hiring a couple of black hats to remotely execute the Mastra workflow on Paul's machine, but settled on something slightly more scalable.
Deployed on Mastra Studio, anyone can simply tag @Corey with a version number & he will generate a draft of the release notes for our review, on GitHub directly. Now if Paul is on vacation, out sick for the day, or fighting to survive a post-apocalyptic London in the grips of a 'Last of Us'-style thunderdome, our release notes process can keep calm and carry on. Automating release notes is nothing to write home about from a technical perspective, but not every workflow augmented with AI agents has to be revolutionary. As in life, if you focus on just being 1% better as an organization every day, you will be astounded at the level of growth you can achieve through compounding returns.
That one win can snowball, and as it turns out so did Corey's responsibilities on the team. When our existing process for processing 'Contact Us' form submissions broke, Corey took over that process as our product marketing agent. As I write this, he is training to handle research industry sentiment analysis on X & has a video-generation skill using Remotion in beta to add some snap, crackle and pop to our content marketing. With the foundation deployed, extending their functionality is a breeze.
El Kapatan — Scaling how we listen to open-source
Our internal agents live in a shared repository on GitHub, where anyone can hop in to extend an existing agent or create a new one from a template. Having documentation of how to integrate to our core data connectors, such as Linear, Notion & Slack, makes it quick and easy for a new contributor to get up and running. Just a few days ago, a new agent started sailing the open waters of open-source after what felt like only an hour of development time - El Kapatan.
Mastra uses a tool/platform called Kapa to power our Q&A 'Ask AI' page on Discord, where users from all over the world ask questions ranging from 'how do I connect local models to Mastra?' to 'wen token?'. It works well for us, and Kapa has a great dashboard with analytics breaking down who asked what, how often, etc. However, it's a classic data silo where if you don't have a login, you would have no idea how to get that data. Engineer and resident comedian Daniel setup El Kapatan so that anyone within Mastra could easily learn more about what our open-source community is looking for.
Bringing this data into the open, into a public chat, creates the opportunity for teams across the company to make more data-driven decisions. Product marketing can generate ideas for content to answer frequently asked questions, or address framework functionality that is poorly understood. Product management and engineering leadership can better prioritize the product roadmap in response to the organic demand bubbling up from the grassroots level. The Customer Engineering & Open-Source teams can more intelligently prioritize which bugs & minor feature requests get attention.
Just get 1% better
Bringing agents into our Slack workspace as publicly-accessible teammates has made internal data more accessible, automated previously manual processes & created a flywheel effect where engineers and non-engineers alike are automating their workflows. Outside of these 3, we have others that focus on analyzing Posthog metrics, uploading & processing YouTube videos, generating personalized Slack briefings, answering questions about the Mastra documentation, triaging and analyzing activity in our open-source repositories, with more on the way.
Many of these improvements are small wins, while some have substantially improved communication between teams. You don't need to replace your leadership team with a swarm of Fable agents, or fuse a Mac-Mini running OpenClaw to your frontal lobe to experience the benefits of agents in 2026. If you just focus on finding a way to make your organization 1% more efficient every day, then at the end of the year you will find yourself at roughly 37x growth with compounded returns. Amidst all the noise in AI today, that's the real signal.
