**Severin Hacker** Co-founder & CTO **Natalie Glance** Chief Eng. Officer # Engineering at Duolingo Welcome to the Engineering hub! You’ll get a glimpse of what our engineers are working on and how we approach delivering on Duolingo’s huge, meaningful mission. Together, we (Severin and Natalie) built the Engineering function at Duolingo, and we’re proud of what we’ve achieved so far. Engineering is Duolingo’s largest function — making up half the company! — and includes these disciplines: - **Software Engineering** (Backend, Android, iOS, Web, SRE, Platform) - **AI / Machine Learning** - **Data Science and Analytics** - **Technical Program Management** - **QA** “Learners first” is a motto at Duolingo, and we really embrace it in Engineering. Our culture of extensive A/B testing means we run thousands of experiments each year, and we’re always exploring new technologies to help us build new features and accelerate content creation for our learners. Ultimately, we’re passionate about innovating together to solve complex technical problems — so more people around the world can learn, better. As Engineering at Duolingo grows, we remain committed to the following: - **Meaningful work:** With our big mission comes the ability to own feature launches that impact millions of users, every day. - **Varied work, larger scope:** Our engineers tackle large, impactful projects so they’re continually challenged and able to grow in their careers — a big reason why so many of our engineers grow with us. - **DEIB:** We work hard to try to have diversity and representation across Engineering levels and disciplines. We hope Engineering at Duolingo can serve as a role model for our industry. We invite you to explore Engineering insights and updates in this hub, and we hope you’ll consider joining our team. ### Engineering Leadership A couple months back at our company all-hands (which we call Parliament, by the way, bonus points if you figure out why), our CEO, Luis von Ahn said: “There's a Slack bot called DuolingoAI — ask it almost anything and it tells you pretty accurately, in a minute or two. It's pretty awesome. I'd strongly recommend it.” The software engineer behind this is Aaron Wang and over the past year, he’s been adding more and more capabilities to the DuolingoAI slackbot and even open-sourced the core framework here: 👉 https://lnkd.in/epRXmazF The DuolingoAI Slack app is built on two things. The first is Anthropic's Claude Agent SDK for the AI layer, which connects the app to a host of MCP servers. The second is the Slack Bolt SDK for reading and responding to Slack messages. The Slack app can perform various read operations on its own and perform human-in-the-loop write operations, like creating PRs or modifying JIRA tickets. It can be triggered by people manually or from other Slack workflows. To date, it now includes 300+ tools for over 25+ MCPs. We use it to power incident response, auto-respond in help channels, run quick data analysis over A/B experiments, open pull requests with human-in-the-loop, list team members on PTO and so much more. Here are more details for two of these examples: - Help Desk Auto-Responses: the DuolingoAI monitors 20+ help desk Slack channels and automatically responds when it thinks it knows the answer to the person’s question. - Incident Triage: when a PagerDuty alert fires, the DuolingoAI investigates using Sentry and Honeycomb, and suggests mitigation steps. It also identifies domain experts in case the on-call engineer needs to escalate. If you want to work at a place where improving developer productivity is a first-class engineering priority, we’re hiring! https://lnkd.in/eMjX9piD #### Learning with AI At Duolingo, we dedicate two hours company-wide every Friday to learning how to use AI to change how we work. What I’ve done with some of that time is shadow software engineers to observe their use of AI coding tools. It’s been very eye-opening. Here are some of the things I’ve learned so far. > One experienced software engineer has challenged himself to not write any code himself at all. “Writing code is 90% solved. LLMs are very, very good at this stuff. What I can do as an engineer is focus on high-level architecture and use my intuition to guide things.” He described AI coding like working with a "super genius intern." He provides occasional hints while leveraging his expertise in high-level architecture, his intuition, and his knowledge of the codebase. > > An intern noted that interns and new grads should prioritize learning foundational skills first. Relying heavily on AI for implementation hinders deeper understanding of intricate system interactions. She uses AI more for explanations rather than direct implementation to enhance her learning process. > > Critical thinking is very important. “Vibe coding” is amazing for unlocking prototyping and tool building for non-software engineers. Software engineers still need to apply their skills to guide AI tools. > > There’s no single front-runner for AI coding tools. Engineers that are successful in using AI have figured out which tools and which models are good for which task, whether it’s debugging a stack trace, fixing a bug, building a new feature, refactoring, migrating code, understanding a repo, etc. > > Tech specs are more important than ever. In fact, good prompting looks a lot like a tech spec. While use of AI coding tools like [Cursor](https://www.linkedin.com/company/cursorllc/) and Claude Code have taken off, it’s clear that we’re still very much in the learning phase. For all the note-worthy AI wins, there are also the AI failures that people are less likely to talk about: going down a rabbit hole trying to solve a problem with AI assistance and then having to restart from scratch the old way. We’re not yet in the stage of seeing meaningful productivity gains that translate into faster product iterations. And that’s okay. It takes time to learn new ways to do work, especially when the tools themselves are changing so quickly. ### Recent Press - [How We Created a High-Scale Notification System at Duolingo](https://www.infoq.com/presentations/on-demand-notification/) InfoQ - [How Duolingo’s AI Learns What You Need to Learn](https://spectrum.ieee.org/duolingo) IEEE Spectrum - [Duolingo’s secret to cost-efficient engineering](https://www.opslevel.com/podcasts/mastering-finops-duolingos-secrets-to-cost-efficient-engineering) OpsLevel