惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Last Week in AI
Last Week in AI
GbyAI
GbyAI
Apple Machine Learning Research
Apple Machine Learning Research
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
N
Netflix TechBlog - Medium
量子位
aimingoo的专栏
aimingoo的专栏
D
Docker
F
Fortinet All Blogs
T
Tailwind CSS Blog
V
V2EX
D
DataBreaches.Net
Google DeepMind News
Google DeepMind News
MyScale Blog
MyScale Blog
G
Google Developers Blog
罗磊的独立博客
MongoDB | Blog
MongoDB | Blog
Microsoft Security Blog
Microsoft Security Blog
Microsoft Azure Blog
Microsoft Azure Blog
WordPress大学
WordPress大学
博客园 - 三生石上(FineUI控件)
小众软件
小众软件
A
About on SuperTechFans
IT之家
IT之家

Mintlify Blog

22 UX improvements to the web editor Introducing the Mintlify Help Center Starter Kit Introducing the collaborative editor built for teams and agents Is your documentation agent-ready? Mintlify raises $45M Series B led by Andreessen Horowitz and Salesforce Ventures 5 things you didn't know you could do in the Mintlify web editor The improved Mintlify CLI Docs on autopilot: From zero to self-maintaining with Mintlify The state of agent traffic in documentation (March 2026) How we built a virtual filesystem for our Assistant We Replaced Our Internal Wiki With a Slack Bot. You Should Too. 8 ways teams use Mintlify to keep docs updated automatically Documentation is your AI interface What three years of watching AI in production taught us Bridging two JSX runtimes: How we solved Astro's React children problem AI agents are shipping faster than anyone can document Knowledge management systems for technical teams Workflows: Automate documentation maintenance Mintlify acquires Helicone to redefine AI knowledge infrastructure Why more product managers are switching to Mintlify Auto-generating documentation sites from GitHub repos Your docs, your frontend, our content engine Take control of your documentation system Almost half your docs traffic is AI, time to understand the agent experience @mintlify for better docs, faster Mintlify for Enterprise Real llms.txt examples from leading tech companies (and what they got right) Mintlify + Claude Opus 4.6: Powering AI-native knowledge management Declaring Clankruptcy: An experiment in agent orchestration Analytics for AI and agent traffic
Workflows, rebuilt
Patrick Foster · 2026-04-29 · via Mintlify Blog

When we launched workflows in March, we gave teams the ability to define automations as version-controlled Markdown files in their project repos. Write a prompt, set a trigger, and the agent handles the rest.

The idea was sound, but the execution left a lot to be desired. We learned from early users and today we're introducing the next version of workflows.

The original design of workflows prioritized flexibility. You could write any prompt and set any YAML frontmatter. Starting with an empty file is the ultimate flexibility since you can take it any direction, but this was also the friction.

Writing a good workflow prompt that produces reliable output takes iteration. You need to test it, improve it, and measure the results. We are not trying to add another maintenace task. We want to save you time and automate away the tasks you're not excited about so that you can focus on other things.

So we looked at how teams were finding the most success with workflows. Two use cases came up over and over:

Keeping content in sync when code ships. When a PR merges in a product repo, something almost always needs updating. Teams were using workflows to close that loop automatically, so content didn't lag behind product changes.

Generating changelogs. The work of combing through merged PRs, filtering out info that's important to users, and writing a coherent summary is exactly the kind of task that's easy to skip and painful to catch up on. A recurring workflow that handles it once a week is valuable to anyone who doesn't find joy writing a changelog.

Beyond those two primary uses, we saw teams running grammar checks, fixing broken links, keeping translations in sync, and auditing SEO metadata. But there was a consistent pattern of teams trying to solve the same specific problems.

So we improved the templates for those common tasks and made them easier to setup.

Today we're launching a rebuilt workflows experience, and it's opinionated.

Instead of asking you to write and maintain prompts, we've built focused, tested workflows for the use cases that show up most. Toggle on a workflow, choose when it runs, and you're set.

Codebase workflows:

  • Codebase updates — Updates content when PRs merge to your product repo.
  • Changelog — Drafts a changelog entry from recent product updates on a recurring schedule.

Maintenance workflows:

  • Translations — Keeps configured languages in sync.
  • Broken links — Finds and fixes broken links.
  • SEO — Audits titles, meta descriptions, headings, and canonical tags.
  • Grammar — Catches typos, spelling mistakes, and grammar errors.
  • Brand tone — Enforces your style guide's voice, tone, and rules.

Each workflow has sensible defaults built in. We recommend triggers we've seen teams have the best results with and update modes based on what the workflow automates. You can override any configuration if your setup calls for something different.

You can also still create custom workflows for anything specific to your project. We've just made the most common tasks easier to start automating right away. Any custom workflows that you've already created are still supported.

We want to reduce the maintenance burden

When you write a workflow prompt yourself, you've got to make sure it's giving you good results. Models improve, edge cases surface, and the prompt that worked six months (or six weeks) ago might not be the one you want today.

With built-in workflows, we own that work. As the underlying agent improves and we learn more from how these workflows run in production, the quality improves for everyone automatically.

Open the Workflows tab in your app and toggle on whichever automations will improve your content.

The first time you enable a workflow, you'll be prompted to configure the trigger and how updates should be applied. After that, it runs on its own.

Go take something off your to-do list and let us know what you think.