Not every decision is the same. Some can be automated, some should be prepared by AI and approved by a person, and some belong to leadership. Here is a ladder.

The short answer

Fashion teams should automate decisions that are routine, rule-based and easy to undo, let AI prepare decisions that need judgement for a person to approve, and keep trade-offs between values, such as margin against sustainability, with leadership. As a 2026 conversation on The Interline put it, there is decision automation and there is decision support, and most of the value today is in doing support well.

Key takeaways

  • Separate decision automation (AI acts within rules) from decision support (AI prepares, a person approves).
  • Automate only what is routine, rule-based and reversible.
  • Most "decisions" are really gathering context; AI can do much of that well.
  • Give every agent a named human owner, and put permissions in the data, not in each app.

What is the difference between decision support and automation?

On The Interline's podcast How Much Will Fashion Let AI Decide?, Rupert Schiessl of Bamboo Rose drew a line that is useful for any product team. Decision automation means the system acts on its own when the rules allow. Decision support means AI prepares a decision for a person to validate. They need different controls, and confusing them is how teams end up either over-trusting AI or getting no value from it.

He also made a point we agree with: a large share of what looks like decision-making is actually gathering context. Which styles use this fabric? Which supplier has tested this trim? What did the buyer say about this colour last season? Getting the right facts in front of the right person is most of the work, and it is work AI can do well when it can reach the data.

Which decisions belong on which rung?

What rules make automation safe?

  • Reversible. If the action cannot be undone cheaply, it belongs on a lower rung.
  • Rule-based. The rule should be written down and owned by a person, not implied.
  • Owned. Schiessl described attaching every agent to a human who is responsible for it. We think that is the right default.
  • Permissioned in the data. He also argued that permissions belong in the data layer rather than in each application, so an agent cannot see or change what its owner could not.
  • Logged. Every automated action should leave a record anyone can read.

Which decisions should stay with leadership?

AI cannot settle a conflict between values. If a cheaper fabric lowers margin pressure but raises the footprint, or a faster supplier costs more, someone has to decide what the company cares about more. The risk with agents is that these trade-offs get made silently, because the objective was never stated. Write the priorities down before you automate anything that touches them.

The Interline's essay Arm's Length AI makes a related point from its survey: teams struggle to say what success looks like for AI at all. Clear objectives are a precondition for both trust and automation.

Where should a team start?

Start on rungs one and two. Retrieval and drafting are where AI saves the most time with the least risk, and they build the record that later decisions depend on. Move a job up a rung only when reviewers have stopped finding problems at the rung below.

What does this look like in practice?

An illustrative example. The brand and the details are invented to show the idea; they are not a case study.

A brand introduces an agent into its product team and sorts its jobs onto the ladder. Retrieval goes in first: the agent answers questions such as which styles use a recycled polyester that has gone up in price. Drafting comes next: the agent prepares the first version of a line sheet and a tech pack for review.

One job moves to automation after a season of clean results: when a colourway is approved, the agent regenerates the listing images for that colourway and queues them for a quick check, with an undo. Pricing stays at the recommendation rung, because a price is hard to reverse once buyers have seen it. The decision to switch a hero fabric to a lower-impact but more expensive option stays with the product director, because it trades margin against sustainability.

Where CotonAI fits.

“Delegate the work. Keep the decisions.” is how we describe the CotonAI Agent. It does delegated work with the right context, from research and drafts to imagery and documents, and brings it back for approval. Approvals are recorded on the style, every access to a workspace is logged, and nothing ships until the right person signs off. That is decision support by default, with people choosing what to automate. See the AI fashion agent.

Sources.

We read these pieces and wrote this analysis in our own words. Figures are credited to the source each article gives; vendor and company figures are their own claims.

  1. The Interline, How Much Will Fashion Let AI Decide?, Ben Hanson with Rupert Schiessl of Bamboo Rose, 1 October 2026. Podcast
  2. The Interline, Arm's Length AI, Ben Hanson, 18 August 2026. Essay drawing on The Interline's 2026 AI survey

Common questions

What decisions can AI make in fashion?

Routine, rule-based and reversible ones, such as regenerating listing images after an approved change. Decisions that need judgement should be prepared by AI and approved by a person.

What is decision support?

AI gathers context and prepares a decision, and a person validates it. It differs from decision automation, where the system acts on its own within rules.

Should every AI agent have an owner?

Yes. Attaching each agent to a responsible person keeps accountability clear and limits what the agent can see and do.

Which decisions should never be automated?

Trade-offs between company values, such as margin against sustainability, should stay with leadership.

Keep reading

Delegate the work. Keep the decisions. We'll show you the workflow end to end and recommend where to start. Book a demo