How the marketing team at Workstream built a fleet of Slack agents the whole team relies on
Jaron Hall runs growth marketing at Workstream. With MintMCP, he wrapped HubSpot's full API in a custom MCP and stood up six agents that live in Slack channels, from marketing ops to deal desk. We talked with him about building AI once so the whole team benefits.
AI that never joined the team
“When an agent remembers last week's context and can act right in the Slack channel my team already works in, it starts to feel like a teammate. That's a real shift in how we work, and it's why we're excited to be building it with MintMCP.”
Lei Xu
Cofounder, Workstream
Workstream builds software for the hourly workforce. Its mobile-first platform handles hiring, HR, payroll, and compliance for the businesses that run on hourly workers, from restaurants to franchises. Like a lot of teams, its people started using AI the obvious way: each person in their own ChatGPT, on their own machine.
That works for one person, and stops there. Nobody could see how a teammate got a good result, and when someone left, the way they used AI left with them. People retyped the same prompt by hand, over and over, with none of the shared context or connectors that would make it dependable.
“The best AI investment you can make is something where you build it and then everybody can benefit.”
Jaron Hall
Growth Marketing, Workstream
Jaron wanted the opposite: expertise built once by the person who knows a job best, then shared with everyone. And he wanted it where people already work, in Slack, not buried in a private chat.
A fleet of coworker agents
Jaron built on MintMCP, and put every agent in Slack channels rather than DMs, so the whole team can see the work and search back through it. The flagship is the marketing ops agent, “mops.” HubSpot's native MCP was mostly read-only, so he used Mint to build a custom MCP that wraps all of HubSpot's API endpoints, then connected the agent to Salesforce, HubSpot, that custom MCP, and Notion.
“The custom MCP to wrap up all those API endpoints was surprisingly easy. I'm in marketing. I've never done this before!”
From there the fleet grew to six agents in production: Scout, which researches news and content topics; a referral program agent that answers reward questions; Sanju, a marketing analytics agent wired to Snowflake and Hex; and a deal desk agent that reviews quotes and loops in finance and revops. The team works with them every day in Slack, and some don't wait to be asked: Scout runs on its own each week and comes back with content angles and outbound signals for the SDRs.
The team runs the whole thing like an engineering group. Ideas go on a Kanban board, get a short PRD, and are ranked on whether they save time or generate pipeline. Jaron even built the product roadmap into the agent as an HTML file it re-checks every Friday.
“Mint is the place where you can build something one time for your whole team to benefit from, and have full control and visibility over how it's actually getting used.”
What Workstream told us
- BuildingWhy build shared agents instead of letting everyone use their own ChatGPT?
- “That's where everybody started, just prompting on their own ChatGPT app. But nobody should be copying the same prompt into ChatGPT or Claude over and over; the agent should hold that context for you, pre-filled and connected to the right data. So we look at how everyone is using their own Claude or ChatGPT and wrap the best pieces into an agent that outperforms all of them individually.”
- BuildingWhy do the agents live in Slack channels instead of DMs?
- “Slack is the place that we're all working; it's just the reality of it. You want your agent to be where the human would have been if the human were doing the work. It's also easier to search the history: if a problem was surfaced four months ago you can find it, whereas that's lost in a DM. Our agents are actually set up so they're only accessible in channels.”
- BuildingWhich agent earns its keep?
- “The one we use the most is our marketing ops agent. HubSpot's native MCP is lacking, mostly read-only. So using Mint, we built a custom MCP that wraps all of HubSpot's API endpoints. Now the mops agent is connected to Salesforce, HubSpot, that custom MCP, and Notion, with a ton of context on how we actually run marketing ops. It handles lead-route troubleshooting, setting up webinars, cloning landing pages and workflows, even just adding a new hire to HubSpot.”
- BuildingYou'd never built anything like this. How hard was the custom MCP?
- “Surprisingly easy. Like I said, I'm in marketing, I've never done this before. I couldn't have done this three years ago; I'd have had to go bug somebody on the engineering team because I didn't know how to make use of an API. Now you point the agent at the API documentation, tell it to always treat that as the source of truth, and ask what's the best way to do this and what are the trade-offs. It walks you right through it.”
- RolloutWhat else have you built?
- “A referral program agent anybody can ask about our referral terms and reward eligibility. One called Scout that searches the news each week for topics we can create content about, and for signals our SDRs can act on. A marketing analytics agent we call Sanju, connected to Snowflake and Hex for reporting questions. And a deal desk agent that takes a look at quotes, evaluates them, and flags the right people in finance and revops.”
- RolloutHow does a new agent idea get built?
- “We try to work like an engineer. We have a Kanban board where anyone on the team can submit ideas, and we evaluate them on impact: is it going to produce more pipeline, is it going to save us time? People write a PRD for their idea so I can review whether it's possible and what the impact would be, and then we get to building.”
- ImpactWhat's the impact been so far?
- “A lot of what I've built is to save our team time on the repeatable things in HubSpot or Notion. Once we build out the full vision it can save some folks up to 30, 40, maybe 50% of their time. It's an investment, because it compounds: you feed it how you operate once, and then it learns more and more.”
- ImpactWhat surprised you most?
- “I love it when you ask an agent a question and it gives you back a better answer than you were even expecting. That happened to me just today with the referral agent: it came back with the answer and a suggestion I hadn't asked for. And it's happening in the channel where everybody can see it.”
- ImpactHow would you describe Mint to a peer in marketing?
- “Mint is the place where you can connect your AI model to your MCP connectors and your custom MCPs in such a way that you can build something one time for your whole team to benefit from, and you can have full control and visibility over how it's actually getting used.”
Move from one laptop's workflow to the whole team's agent
See how MintMCP lets one person build an agent the whole team can use. Read more stories