Claude — searchable YouTube transcripts & AI Q&A

Coding on the go with Claude Code

Claude Code's cloud sessions let you start a coding task on your laptop and continue it from your phone or another device, even while your computer is asleep. The session runs on a virtual machine that clones your repo from GitHub, and you can review changes, leave comments, and create pull requests from your phone. This is useful for long-running tasks like refactors or migrations, but it requires giving Claude access to your GitHub repos and doesn't include your local settings or plugins.

Getting started with Claude Code Projects

Claude Code Projects lets you hand a large goal to Claude, which breaks it into parallel threads that run in the cloud and can continue even when you close your laptop. You can steer threads, have them open pull requests and fix CI failures, and set up daily routines to catch regressions. The speaker notes that threads use plan limits faster, so you can adjust model and effort settings, and that threads share memory for consistent preferences.

Inside Shopify's AI strategy with CTO Mikhail Parakhin

Shopify's CTO argues that AI's biggest impact is raising the ceiling of what's possible, not just automating routine tasks. He illustrates this with a system that models a company as a sequence of actions, enabling counterfactual simulations to guide merchant growth strategies. He advises other CTOs to always use the largest models for development and to fine-tune smaller models for production to balance cost and quality.

Building secure agents for knowledge work

The speakers argue that AI agents in the workplace are most effective when they act as a shared coworker, learning and spreading an organization's taste and best practices. They illustrate this with their 'Kate Bench' skill, which encodes an editor's historical edits so the agent can apply her style, and they emphasize that such systems compound expertise, allowing experts to scale their impact without spending more time. They also note that while benchmarks are useful, the real measure is how well an agent performs on your specific work, and they advocate for building personal benchmarks from your own examples.

Mods in Claude Code: change how Claude Code works

Mods in Claude Code let you directly alter how the tool behaves by writing TypeScript functions that hook into internal events, acting as middleware. For example, you can hide email addresses in replies until hovered, while the model still uses the real addresses. Mods are plugins with an extra `modules` key in `hook.json`, run in a sandbox, and can be shared via the marketplace.