Help Scout is rebuilding design work for an AI that can actually use it

Help Scout has adapted its design process for AI by making its principles explicit, rebuilding its design system so language models can interpret it, and creating internal tools for prototyping and research. The practical result, according to designer Buzz Usborne, is faster execution and more production code without lowering the team’s quality standards.

Watch Desk analysis

What happened

In a post published on 22 September, Usborne describes how Help Scout changed its design workflow around AI rather than simply adding a chatbot to the side of it. The company wrote down design principles, made its design system more machine-legible and built internal AI tools for prototyping and research.

That machine-legible point is the quietly important one. If an AI system is expected to work inside a design process, it needs more than screenshots and vibes. It needs structured guidance about components, rules and intent, so its output has a fighting chance of fitting the product.

Why it matters

The claim is not that AI has replaced Help Scout’s designers. It is that the surrounding system has been redesigned so people and language models can collaborate with less friction. Usborne says this helped designers work faster and ship more production code while maintaining quality standards.

That offers a more useful picture of workplace AI than the usual promise to automate an entire profession by lunchtime. The leverage appears to come from documentation, tooling and a design system that can be understood by both people and software. Tedious preparation, it turns out, remains remarkably employable.

Our read

This is a credible account of AI adoption because it describes the unglamorous work required before the flashy bit becomes useful. Explicit principles and machine-readable systems may not make for a thrilling launch video, but they give AI something better than a blank page and a prayer.

The important caveat is attribution. Usborne presents the speed and quality gains as the outcome of Help Scout’s changes, but the supplied account does not provide independent measurements, comparative benchmarks or details of how quality was assessed. Teams considering a similar approach should borrow the method, then measure whether it actually improves their own work.

What to watch

• Quality measurement:
  whether Help Scout publishes concrete evidence for its claim that faster delivery did not reduce quality.
• Design-system structure:
  which standards and metadata make the system useful to language models.
• Human review:
  how designers check, correct and approve AI-assisted production work.
• Wider adoption:
  whether this process spreads beyond design into research, engineering and customer-facing work.

Discussion spark: Is the real advantage of AI in design coming from the model itself, or from the painstaking work of making a company’s systems legible enough for the model to use?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.