Jam.dev

Request/response body + all headers in every bug report.

⭐️⭐️⭐️⭐️⭐️ Thanks for the kind review! 💜 #ChromeReviewBot

We hosted 100 devs for AI talks w/ Sourcegraph and Cloudflare! Watched back the clips with Ivanha Paz and discussed! Here's the future of AI from Tejas Kumar, Beyang Liu and Zach Bai 🍿 Preview here, YT link below!

This week we got to host tech talks on the future of engineering w/ AI 🔥 The talks were so good - really SO good. We're excited to share the highlights! Now live on the pod. S/O to our awesome cohosts Sourcegraph & Cloudflare! And to the brilliant speakers - y'all crushed it! Tejas Kumar, Zach Bai, Beyang Liu

Jam.Pizza on the way to builders in Vancouver for Techstars Startup Weekend! Hosting dev meetups? The pizza's on Jam! Apply at Jam.Pizza

We did the math, reporting bugs takes too long Cut all the steps down to 1 click - next time you see a bug just Jam it! ⚡️ #bugreporting #QA #devtools #debugging

Write automated tests from Jam’s auto-generated repro steps! Just grab the data-test ID from component attributes.

Building AI is super fun! But... not easy. I gave a talk on this topic at Cloudflare HQ last night. Here's 5 lessons building AI at Jam.dev + 5 tools we use: 1/ Bad AI is worse than no AI The quality bar for AI is super high. When you use a bad AI feature, it breaks user trust, it feels like the company prioritized hype > experience. So, better not to ship AI at all than to ship bad AI. We did a 1 week hack week, then made a go/no-go decision whether to continue building our AI feature. We'd rather not ship at all than ship a bad AI. 2/ The AI is either good, or it’s not. No in-between. We had to make a decision – should there be an AI button users press to get an AI-generated Jam description, or should it happen automatically? But we realized, either the AI is good – in which case it should auto write all your Jams, or it's not, in which case we shouldn't ship it at all. There's no in-between "let the user decide if they want AI" state. 3/ Build systems to get user feedback AI is as good as your user inputs. We setup multiple input sources for feedback, and review them as a team every Monday: 1. Data: % AI accept rate 2. Feedback: % thumbs up rate 3. Qualitative: user emails + calls 4/ The “magic vibes” days are over Lots of AI products have that magic sparkle vibe to show "magic AI stuff is happening" But if you believe AI is just a utility and the way of the future, it should actually blend in No clutter, no magic, it's just the new way tools work. So we decided, no special sparkle emojis, and pink & purple gradients. It's greyscale and blended in, just plainly part of how Jam now works going forward. 5/ Here are the tools we use for prompting! Promptfoo, Anthropic, Glif, Notion, Metabase Shoutout to the boring tools. Nothing can replace a great Notion log of prompt experiments tried, or a Metabase for tracking AI product quality. -- Was a super fun eng AI talk event! Shoutout to our cohosts Sourcegraph and Cloudflare! See you at the next one.

Why Copilot is our favorite AI coding assistant ⚡️ It makes us 20% faster at writing code - especially when naming variables & writing comments.

Presenting now at AI Eng Talks: Tejas Kumar ⚡️ The future is smaller language models with fewer parameters - like a specialist! So it gets one thing right, every single time.