I just completed my routine 'CEO does Support' rotation. If you got a reply from 'Martin', it was me 😀 Some cool things I got to help with, pain I got to simmer in, and the answer to the question of whether AI can replace support teams (in comments): 1️⃣ A customer wanted a 'vibe check' on how to implement an Age Gate that preserved SEO. i.e.: normal users get presented with the Age Gate, but Google goes through. When they implemented it naively, SEO tanked. I went to v0 by Vercel and created a demo on how to implement this with Middleware, then contributed to Community: https://vercel.fyi/gate (press 'Built in v0' to access the template). (Note: while 'cloaking' is usually frowned upon, Google makes exceptions for age gating for legal reasons: https://lnkd.in/gwhwkVJS) 2️⃣ A customer from a large consulting firm was having issues with the deployment pipeline. It was a high priority ticket due to the SLA we honor and criticality of the project. It was an unusual edge case related to exceeding function size from a large dependency. The "build" phase was fine, but the "deploy" phase was not. I recommended two things: ▪️ Little known fact. Vercel CLI has a command called 𝚟𝚎𝚛𝚌𝚎𝚕 𝚋𝚒𝚜𝚎𝚌𝚝 that performs a binary search so that you can try out Deployment URLs and find the one that introduced the regression. It's a life-saver ▪️ Try to identify accidental importing of large deps, look at the output of 𝚟𝚎𝚛𝚌𝚎𝚕 𝚋𝚞𝚒𝚕𝚍, and use the bundle analysis tools in Next.js Takeaway: we have an opportunity to do better in the logs view and help you explore the inputs and outputs better. We're also working on Vercel Agent automatically helping diagnose issues and open PRs, which should help in the future. 3️⃣ A customer was having issues understanding the lifecycle of Image Optimization. They updated an image on their DAM (place where they host images), but the optimized cache copy was not reflecting a change. This uncovered a few issues: ▪️ The TTL behavior is hard to understand. We respect 𝙲𝚊𝚌𝚑𝚎-𝙲𝚘𝚗𝚝𝚛𝚘𝚕, but there's also a 𝚖𝚒𝚗𝚒𝚖𝚞𝚖𝙲𝚊𝚌𝚑𝚎𝚃𝚃𝙻 param. We can improve docs and observability. ▪️ DAMs should be able to trigger a preemptive revalidation of our cache. Customers shouldn't have to wait. We're working on the CDN APIs to tell Vercel "recompute this value in the background, and only replace the cache if it works and is error-free". This is basically what Next.js ISR does and it's a godsend, but we're building APIs for it. One of the common reminders from Vercel scale: there's no such thing as too much observability. There's endless appetite for logs, traces, and more ways to inspect the behind the scenes of the machine. 4️⃣ Our support team is absolutely cracked... (continued in comments)
KTLO
Such a great way to stay in touch with your users. Keep it up, Martin! 🤣
Somewhere out there is a customer bragging “I got debugged by Martin” and no one believes them.
Hi Martin
This is great, Guillermo Rauch I believe this should be set as a standard practice in the industry—that every CEO should wear the support hat at least once a year. This will surely provide an amazing perspective on the real challenges customers face along with some unique use cases and create stronger alignment between leadership and frontline teams.
Just wish that rollback functionality actually worked. After being rolled out earlier this year, it worked for months and was an efficient guard against bad changes. Then it stopped during the summer. Huge PitA.
Always wild how much you learn doing support shifts - real world's messier than any dashboard.
Gran elección de nombre falso.
CEO at Vercel
4d4️⃣ Our support team is absolutely cracked, and I'm extremely thankful for their dedication to our customers. I spent quite a bit of time debugging how we can make our own support experience better with AI. Even though we already answer tickets with AI, and even though we already have a 'AI support assistant' for the behind-the-scenes, there's still huge amounts of alpha. Takeaways: ▪️ A lot of the support AI tools were conceived as one-shot assistants. What we've learned with v0 by Vercel is that there's a 100x upside in upgrading from "models" to "agents". ▪️ We're working on content drafting agents that allow us to take common patterns and best practices and create durable content from them. Great opportunity for dogfooding our AI Cloud capabilities. ▪️ Support Engineers are absolutely crucial and AI is not replacing them anytime soon ☺️. AI is getting rid of the uninteresting tickets. The ones that remain require significant effort and creativity to tackle, and, crucially, inform product roadmap and decisions. I expect AI to continue to have an enormous impact here. We're not nearly 10% done accelerating delivering quality help to users, and there's a lot more we can do to help them help themselves.