
A console.log in a commit is embarrassing. An API key in one is worse. Ship Gate is an agent skill that scans your staged diff for secrets, debug code, and stray files, then gives you a straight PASS or BLOCK.

Built for developers who want to unlearn outdated defaults and start building engineering judgment. Powered by @Bitter_Brains

A console.log in a commit is embarrassing. An API key in one is worse. Ship Gate is an agent skill that scans your staged diff for secrets, debug code, and stray files, then gives you a straight PASS or BLOCK.

The free Vue.js exam has left the bundle. The $149 hasn't 💚 Until Tue 24 Nov, 22:00 CET, $149 still gets you a year of Vue School Premium and Unlearn.dev, plus Mastering Nuxt Complete and Mastering Pinia Complete. 👉 vue.school/vsbf26

Quick question. How many date formatters does your codebase have? One team found 14 near-identical HTTP retry wrappers, each written by an agent that didn't know the last one existed.

Every answer you give is one less thing the agent decides on its own. It's from lesson 3 of Spec a Product Before AI Builds It. Lesson 1 of every Unlearn course is free, no account needed → unlearn.dev/courses/plan-the…

Before an agent builds from your spec, run this first 👇 "Read spec.md and list every assumption it makes about users, roles and data. For each one, ask me the question that would remove it. Don't rewrite the spec yet."

Web Summit Day 2 got cancelled. So we threw our own party. Within 24 hours, we found a venue, promoted the event, and opened registrations. 386 people signed up. Over 200 showed up. No months of planning, no big budget. Just a room full of developers, product and design people, founders and builders who wanted to meet each other anyway. It was one of the best nights of the week. Thanks to everyone who came, shared the invite, and stayed late. You made it. We're at Web Summit Day 3 today. If you want to grab a coffee or a drink at the after-party tonight, send a DM to Jasper Willems. Thank you to Jos Gerards and DEVWorld Conference for co-organising this!

Met a robot at Web Summit. Great posture, very polite, zero stickers. We've got plenty. The Unlearn team is still in Lisbon, come find us and talk shop 👇 calendar.notion.com/meet/jwi… #WebSummit

Stranded in Lisbon after Web Summit? ⛈️ Devworld Conference × BitterBrains are hosting drinks tonight from 7 PM CET at Saratoga. No talks, no pitches, just good people. 100+ going; seats are limited 👇 luma.com/j5k2jekf #WebSummit

MCP still wins when your agent has no terminal, there's no decent CLI, or the agent acts for people with their own logins. The full comparison has replays of every run → unlearn.dev/comparisons/mcp-…

Across eight tasks on Sonnet 5.5, both setups got 24 of 24 answers right. MCP used 2.1× the tokens and cost 2.1× as much. With every tool definition loaded, GitHub's server adds about 20k tokens before you've typed anything.

gh let the agent ask for five fields and filter the reviews with --jq before anything reached the model. The MCP tool sent the whole PR and every review, and the agent read through all of it to find the same three answers.

We asked Claude Code to look up the same GitHub PR two ways, through the gh CLI and through GitHub's MCP server. Both got it right. The CLI got back 135 characters. MCP got back 19,564 🧵

The scariest file on your laptop is the .env your coding agent can read 🔑 Swap in test keys, a staging database and tokens scoped to one repo. It takes ten minutes, and if the agent ever gets tricked there's nothing worth stealing.

It's one of the eight mistakes agents make most when writing code. The other seven are just as easy to miss in review → unlearn.dev/blog/the-8-most-…

Asked an agent to write tests. Got this back, green on the first run. It mocks calculateDiscount, then checks the mock returns 20. The real function could return anything, or not exist at all, and this would still pass. It's a test that Jest works.

Approval prompts are the lightest of four ways to limit what your agent can reach. The other three are sandboxes, dev containers and a separate machine → unlearn.dev/blog/how-much-ac…