Sene extends its Shopify operations platform with Claude and human merge control

This is what a Shopify development agency does once AI can write the code: the judgement, the platform knowledge and the decision about what enters the shared branch. Sene Studio had an operating platform before Claude entered this story. Autonomous had spent years developing the connection between Shopify, custom-order workflows, factory integrations and an operations console. Sene’s founders were contributors to that platform, with direct access to the code and the knowledge needed to shape its next changes.

By mid-2026, the relationship was changing. The founders were becoming more self-sufficient with AI, and storefront responsibility moved to Ray Li. Autonomous continued AI-related support. This case follows the observed Claude-authored work on the platform and dashboard, where a feature can affect a real order and the person responsible for fulfilling it.

The changes were recognisable business tools. Dashboard work included SKU generation, a searchable generation log, product and fabric filters, and search by tracking information. These capabilities help staff find, create and inspect the records they already use to run the business.

Other changes addressed the handoff between the platform and manufacturing: handling a changed shipping address before an order is fulfilled, making fit-update failures visible in the error queue, and recording an identifiable reviewer for a fit update. Their value comes from the operational decision they support, not from the fact that an agent wrote code.

The scope crossed a Django backend and a Vue operations interface. That matters because a usable feature often needs both the API behaviour and the controls an operator sees. A dashboard button alone cannot establish that the underlying workflow does the right thing.

Workflow: Operational task → Claude-authored change → Human merge decision → Platform capability.

No change reached the shared code because an AI wrote it. A named person accepted it. The observed agent-branch changes entered the shared development branches through named human mergers. Mark Zheng, Sene’s co-founder, performed most of those merges; Autonomous engineer Danial handled others. The record supports a clear statement about merge control.

It does not establish that every merge included the same depth of review. A merge record identifies who accepted a change into the branch. An explicit review-response change shows something more specific: a problem was raised and the implementation was revised.

The repositories contain concrete examples of substantive review feedback followed by corrections. They also contain human-authored work on branches carrying the same naming convention as the agent work. Authorship must therefore be checked directly rather than inferred from a branch label.

In plain terms: the AI wrote tests for its own work, and that is useful, but it is not the same as someone asking whether the feature should behave that way at all. Agent-authored work included tests alongside implementation. Those tests provide examples of expected behaviour and support future changes, but their presence cannot establish how the system performed in production. Human review remains responsible for asking whether the behaviour is appropriate for the workflow in the first place.

The practical review surface is broader than whether a feature appears to work once. Exports need to contain the intended records, actions need to respect order state, and the interface needs to match the backend contract. The observed corrections show why an established platform benefits from engineering judgement even when code creation becomes easier.

This is a collaboration between an existing business system, client knowledge, AI-authored changes and people who decide what enters the shared branch. Each has a distinct role.

By September 2026, client and agent contributions were carrying recent platform development. The commercial result described by Autonomous’s founder is greater client self-sufficiency: Sene can build on the automated workflow foundation with Claude, while the engineering relationship adapts to support that way of working.

This case does not assign all agent operation to Autonomous or claim a measured delivery-speed improvement. It shows an established platform becoming a foundation for client-led development, with human control visible at the merge boundary and substantive corrections visible in part of the review history.