AI Compass / Quick reads
An agent can forget a rule, but not a gate
Better instructions alone do not bring quality. It takes checks the agent cannot get past.
In depth: Checks the agent cannot bypass – checkpoints and gates
Shopify has AI agents work through large undertakings in small checkpoints. Every checkpoint has to pass four gates (checks): first automated tests, then a comparison with the interface, then an adversarial review (at Shopify two independent reviewers who must both agree), and finally sign-off by a person. AI LABS rebuilt this in Claude Code: a stop hook (a script that Claude Code runs before the agent stops) keeps it in the process, and notes from the sign-off land in a learning file for all agents.
Tool and alternative
Helix is internal to Shopify and not available. The skills from AI LABS cost money (AI Labs Pro); according to the video description, you can also do it yourself with a prompt. Open option: according to AgentConn, the rebuild helix-loop (MIT licence, skill package with a stop hook), still early, not tested by me.
Sources
- AI LABS, “Shopify Just Released The Greatest AI Coding Workflow Ever”, 24 September 2026 (chapters 03:01, 07:32, 09:18, 11:56, 13:11)
- Shopify Engineering, “Helix”, 21 September 2026 (primary source)
- Context: AgentConn, “How Shopify Wired Claude Code”, 24 September 2026
Advertising in the source: Video sponsored by Hedra (unrelated to the topic).
Want to try this with your team? That’s what the AI Lab is for. See the AI Lab