Tech • AI • Robotics • Game
Claude Code Projects beta joins Shopify’s merchant digital twin
Briefing
Claude Code Projects enters beta
Anthropic introduced Claude Code Projects in beta as a project-based workflow for long-running software work. The feature keeps a single persistent goal in view while Claude breaks it into smaller tasks and coordinates execution. It is positioned for jobs that stretch across many sessions rather than one-off coding prompts. The launch expands Claude Code from an interactive assistant into a higher-level orchestration layer.
Claude splits work into threads
The core mechanic is parallel task execution through separate cloud threads, each operating like an independent session. Claude can assign subproblems to different branches at the same time, reducing the need for developers to manually juggle sub-agents or worktrees. That makes sprawling maintenance projects easier to structure and monitor. The approach also signals a push toward agentic software workflows with built-in concurrency.
Cloud branches handle ongoing engineering
Each thread runs on its own branch in the cloud, allowing work to continue while the user is offline. That design targets multi-repository efforts where fixes must land across interconnected systems rather than in a single codebase. Anthropic is effectively turning background execution into a native part of the development loop. The result is closer to delegated engineering management than chat-based code help.
Claude can open and track PRs
Beyond writing code, Claude Code Projects is designed to open pull requests, follow review status, and keep tasks moving. It can also run recurring checks, giving teams a way to automate follow-up work after the initial implementation. This shifts the product from code generation toward end-to-end workflow supervision. In practice, Claude becomes responsible not just for drafting changes but for shepherding them toward merge.
Hybrid mode taps local tools
When a task requires local files, developer tooling, or machine-specific context, a thread can run on the user’s computer instead of fully in the cloud. That hybrid setup addresses a common limitation in remote coding agents, which often lack access to local environments. Anthropic is trying to bridge background cloud execution with hands-on workstation tasks. The model suggests a flexible deployment path for real-world engineering teams.
Shopify builds merchant digital twins
Shopify outlined a broader AI strategy built around predicting business outcomes, not just automating routine tasks. Its HST prediction system creates a digital twin of a merchant and models likely results from different interventions. Teams can test alternatives such as loans, faster shipping, or ad campaigns before acting in production. The ambition is to simulate business decisions the way language models predict the next token.
HST runs counterfactual business tests
The HST system is designed to ask what would happen if a merchant changed a specific variable, then estimate which path best supports growth. That means Shopify can evaluate interventions such as financing offers, payment changes, shipping adjustments, or marketing prompts on the model first. The emphasis is on counterfactual reasoning rather than simple dashboard analytics. If accurate at scale, the system could turn merchant support into proactive decision optimization.
Shopify argues AI lifts ceiling
Shopify’s framing goes beyond productivity gains from speeding up known tasks. The company argues AI’s larger payoff is raising the capability ceiling by making previously impractical work possible, including company-level prediction and management foresight. That view contrasts with narrower automation narratives centered only on labor savings. It positions AI as a tool for new classes of decisions, not just cheaper execution.
Videos covered
Inside Shopify's AI strategy with CTO Mikhail Parakhin
- •From text prediction to company prediction
- •A digital twin for merchants
- •Ceiling, not just floor
Getting started with Claude Code Projects
- •Project-based coordination
- •Parallel threads in the cloud
- •Option to use local files and tools