Mintlify is powering agent-native documentation for over 20,000+ companies.
We wanted to know about the AI stack they use to operate internally - what automations let them move fast. So, we sat down with CEO Han Wang and asked him to walk us through every important agent and AI workflow they use inside Mintlify.
Here’s what we heard:
1. Agents that take support tickets → fixes in prod
Mintlify built a system that ingests every customer support ticket, identifies if it’s about an issue, and if so, assigns it to a coding agent to fix.
How it works:
An AI support agent (they use Parahelp) responds to incoming support tickets in their helpdesk (they use Plain)
A Slack integration pushes every new support ticket to Slack, too
A background coding agent (they use Replicas) reads every Slack message and uses the Linear MCP to create tickets for anything actionable
The Linear Agent auto ingests every incoming ticket, and assigns Claude Code to all bug tickets
Claude Code takes a first pass automatically and submits a PR
Two AI code reviewing agents (Cursor BugBot and Greptile) review every PR and give a confidence score
Engineers do the final review, run tests and hit merge when tests pass
Mintlify says this helps the team ship faster and reduce the time between a customer-reported issue and a production fix.
2. Design prototyping
Mintlify’s design team wanted to unblock engineers to run quickly on discrete features so they can ship at the speed of good ideas, and so that the design team gets to spend their time working on more high leverage, longer-term research and work.
They realized a lot of vibe coding platforms are good enough at creating a first pass at designing basic features – enough that a seasoned designer can look at the prototype, point out a couple changes and let engineers run. They set up both Lovable and Replit with the Mintlify design system so that the engineering team can prototype new features quickly in either platform.
When building a new feature, an engineer drops the Slack thread where the feature is discussed into Replit or Lovable for a quick prototype, and send the result to designers for a gut check & any edits.
3. A knowledge retrieval agent
Mintlify also wanted a knowledge agent to unblock anyone in the org when a question arises. So, they built a knowledge base agent with access to Notion + Google Drive + meeting notes + Slack + emails + codebase. When anyone has a question, they just at mention the KB agent in Slack.
It’s a simple idea, but it can make a huge difference in how much deep work gets done. Fewer people get interrupted, and people get unblocked faster.
This one is actually running on their own platform - they offer a shared memory layer that pulls in all internal code and communication and creates a sort of knowledge runtime for internal agents.
4. The CEO’s personal agent
Han built himself a “Han agent”, inspired by personal agents like OpenClaw.
The Han agent has access to all the tools he uses, and is available right in Slack. So instead of switching contexts to read/write to any app, he just asks the Han agent to do things for him right from Slack.
How it’s built: Gumloop + integrations to all the tools.
5. Email agent
This one helps the team spend less time in emails, more time producing work, while still being highly responsive.
It checks emails every morning and marks spam or unimportant messages as read. It auto-forwards any emails that need to be routed to another team member automatically. For every real email, it drafts a response using past emails and the internal knowledge base.
How it’s built: Gumloop + email/Slack integrations
6. Morning sales brief
Every morning, this agent sends a rundown of all customer conversations and in-process deals.
How it’s built: Claude Cowork cron job using MCPs from Salesforce, Granola, and Attention to pull customer conversations and pushes to Slack
7. The weekly intel agent
Each week it compiles what customers, partners, and adjacent companies shipped, then drops the roundup in a channel.
How it’s built: a vibe-coded script on a cron job, calling the Claude API with deep-research tools attached.
The full stack
Coding: Claude + Cursor.
Design: Replit + Lovable.
Personal-agents: Gumloop
Tickets: Replicas (routing to Linear) + Linear agent (assign to Claude) + Claude (create PR) + Cursor/Greptile (review PR)
Orchestration: Slack (interface) and Temporal (durable workflows).
Finally, Mintlify’s agent vs hire framework
Not everything is an agent. The team is still growing and actually 3xed in size last year. Here’s Han’s framework…
For every new problem that arises, check:
Is this a problem that can be solved by automation/agent?
Is this a problem that can be solved by a process change?
If no & no, solve with headcount.
Is there another company you’d like to see the internal AI stack of next? Let us know.
This newsletter is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. Furthermore, this content is not investment advice, nor is it intended for use by any investors or prospective investors in any a16z funds. This newsletter may link to other websites or contain other information obtained from third-party sources - a16z has not independently verified nor makes any representations about the current or enduring accuracy of such information. If this content includes third-party advertisements, a16z has not reviewed such advertisements and does not endorse any advertising content or related companies contained therein. Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z; visit https://a16z.com/investment-list/ for a full list of investments. Other important information can be found at a16z.com/disclosures. You’re receiving this newsletter since you opted in earlier; if you would like to opt out of future newsletters you may unsubscribe immediately.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.