Case Study Club

Welcome to Volume 3

This is Volume 3. Everything's changed.

If you just joined, hello. If you've been here a while, thanks for sticking around.

I'll be honest. I haven't given this newsletter the attention it deserves.

You signed up. You gave me your email in exchange for what Case Study Club promised. I don't feel like I've always delivered on that.

The friction got in the way ; copy-pasting from Webflow into EmailOctopus, hunting for signals, no clear format. It came together randomly.

That's changed. For the past few weeks, I've figured out what I actually want this to be. And I've built a system that's only been possible now.

Every day I get a roundup of updates from my sources. Each Friday I sit down and figure out what deserves extra attention. What's moving in the industry. Then I report back to you with my findings and takes.

All that manual copy-pasting, clicking buttons, filtering, selecting templates ; gone. The friction is gone.

Case Study Club started as a newsletter. Now I'm going full circle.

Reporting on what actually works in the design industry right now. It helps me navigate this noisy world. If it helps you too, even better.

Welcome to Case Study Club, Volume 3.

— Jan

Claude Code as a product problem

Intercom doubled its engineering velocity in nine months by treating Claude Code adoption as a product problem rather than a tool mandate. Brian Scanlan, the company's engineering leader, described the rollout in an interview with Lenny's Newsletter, naming three pieces of infrastructure that landed first: a custom skills library, a telemetry dashboard, and an automated standards enforcement loop.

The rollout reached every R&D function. Engineering, design, product management, technical program management. Each role was expected to operate inside the same AI-augmented stack as a baseline. The skills library codified how Claude Code should be used against Intercom's specific codebase, review conventions, and product constraints. The telemetry dashboard measured adoption and output at team and individual level. The enforcement loop flagged AI-assisted output that drifted from the company's quality standards before it reached a human reviewer.

Treating the rollout this way meant the tool received the same seriousness Intercom would give to launching a feature for an external customer. Internal tooling, in most organizations, runs on a thinner discipline than that. The discipline showed up most clearly in the cross-function reach. Designers and product managers were not adjacent users of the engineering team's tooling. They were inside the same operating system as the engineers, pulling against the same skills library, measured by the same telemetry.

The 2x figure did not arrive on bare ground. Scanlan is direct about what was already in place. CI/CD pipelines were mature. Test coverage was comprehensive. The culture was high-trust, with engineers shipping under low-friction review. Claude Code amplified that base. It did not build it. The honest reading of the velocity claim is not "AI made us twice as fast." It is closer to "our existing infrastructure could absorb a 2x increase in throughput without breaking, and the tool brought the throughput."

The more consequential finding, buried inside the velocity claim, is structural. Intercom's R&D team observed that their own product surfaces now need to be navigable by autonomous agents by default. If an interface blocks AI access, customers build their own workarounds outside the product. That shifts the competitive pressure from user experience to developer experience. The agent becomes a user a designer has to serve, not an infrastructure concern handled downstream by a platform team. Whose intent the interface honors becomes a design decision again, not a default.

The implications for design are concrete. Screen readers needed semantic HTML; agents need structured endpoints, stable identifiers, predictable affordances, and documented surfaces to act on. The team that treats agent interaction as an accessibility class of problem, planned for and tested against, will ship products that still function when the user is a model. The team that treats it as a developer-relations concern will find its product being worked around.

The 2x figure is self-reported through one interview series, with no independent verification this week. What Intercom documented concretely is the infrastructure underneath it: skills library, telemetry, enforcement loop, all on top of foundations that existed first. The infrastructure transfers. The number does not.

A UC San Diego study found 55.8% of LLM-generated ecommerce UI components contain at least one deceptive design pattern.

Anthropic's product team, per Cat Wu, has compressed its shipping cycle from months to weeks to days.

Eye magazine's issue 109 used only two TypeTogether typefaces throughout, Futura 100 and Abril Text.

Microsoft, Google, and IBM each published AI design frameworks: HAX Toolkit, PAIR Guidebook, AI Design Guidelines.

Nikhyl Singhal projects 30,000 traditional PM roles disappear within two years, with 8,000 new AI-first positions in their place.

The slot-machine ads design history forgot

In Play Meter, a trade magazine that ran through the late 20th century covering coin-operated entertainment, Eastern Gaming Tech runs ads through the early 1990s for its token-dispensing slot machines. The machines themselves photograph in the ads, in their own way an aesthetic of the period. But the typography does the heavy lifting. Three fonts. Banco, Sinaloa, and Max set against each other across multiple placements, the font mix carrying the campaign's structure and rhythm.

It reads like a budget tell as much as a craft one. The print shop's available type was the design budget. The juxtaposition that came out is more deliberate than what a richer budget would have produced.

Fonts In Use catalogued these recently. Trade-press work like this often falls outside mainstream design history because the trade press doesn't form the canon. The bulk of working design happened there, and still does. The Eastern Gaming Tech ads are an entry from that pile.

A Unicode browser that thinks visually

Charcuteria, by Elastiq, is an interactive Unicode browser. It treats character exploration as a visual and relational problem, surfacing the connections between related glyphs instead of presenting them as a linear list to scroll through.

For type designers, anyone working with internationalization, or anyone building a symbol system, the interface gives a quick way to see how Unicode's pieces relate. Glyph proximity, script ancestry, structural similarity. The kind of map otherwise reconstructed from scattered tables.

Worth opening the next time a character should exist somewhere in U+ space and the standard search isn't surfacing it.

Case Study Club is written and edited by Jan Haaland and published weekly from Norway.

If this issue was worth your time, forward it to one person who would feel the same.