AI Browsers

Amazon moves to block Perplexity's AI shopping agent

From Casey Newton at Platformer ($):

"Today, let’s talk about how Amazon’s fight with Perplexity previews the next major fight between AI companies and the rest of the web. AI companies entered the year determined to make 2025 ‘the year of the agent’—but ran headlong into institutions that aren’t having any of it."

The battle over AI agents' access to the web is heating up. Amazon sent Perplexity a cease-and-desist letter demanding that they stop allowing Comet—Perplexity's AI browser—from shopping on Amazon.com on users' behalf. Amazon argues that third-party applications should respect whether businesses want to participate, while Perplexity fired back with accusations of "bullying," claiming that Amazon simply wants to control the shopping experience (and serve you more ads). This isn't Perplexity's first clash over web access. Cloudflare previously caught them using "stealth crawlers" to bypass website blocks, and Reddit has sued them over alleged data scraping.

The stakes here are bigger than just shopping convenience. As Casey Newton points out in his analysis, the foundations of today's internet were built assuming humans visit websites and conduct transactions. If AI agents become the primary interface between people and the web, many business models will break. The outcome of fights like this will determine whether we end up with a decentralized web, where different agents and sites negotiate access, or a more consolidated future, where a handful of AI companies control how we interact with everything online. For now, AI agents remain slow and unreliable shoppers—but that will no doubt change.

The Verge | Amazon and Perplexity have kicked off the great AI web browser fight
⚡ Quick Read (2 minutes)

Platformer | Amazon gets hit by a Comet ($)
☕ Medium Read (7 minutes)

AI Product Development

When developers move at 10x speed, how do designers respond?

AI coding agents are delivering measurable productivity gains for development teams. Amazon engineer Joe Magerramov reports his team is now producing code at 10 times their previous velocity—not through reckless "vibe coding," but through what he calls "agentic coding," where engineers collaborate with AI agents while remaining accountable for every line of code. The approach combines human oversight with AI speed. But this dramatic acceleration creates ripple effects throughout the software development process, requiring teams to rethink everything from testing strategies to communication patterns.

For design teams watching developers sprint ahead, Luke Wroblewski identifies three common responses:

• Some are embracing a role shift, moving from creating mockups upfront to ensuring UX alignment after features are built, essentially flipping the traditional design-then-develop workflow.

• Others are picking up AI coding tools themselves, with designers at companies like Perplexity and Sigma now shipping code directly or fixing UX issues in production.

• A third group remains skeptical, viewing AI-accelerated output as "faster slop." But as Wroblewski notes, citing Sturgeon's Law, 90% of everything is mediocre regardless of the tools used. The fundamental challenge of creating quality work hasn't changed, even if the velocity has.

How design teams are reacting to 10x developer productivity from AI
⚡ Quick Read (3 minutes)

The new calculus of AI-based coding
☕ Medium Read (9 minutes)

AI Agents

A primer on context engineering: the framework behind AI agents

Tony Seale's recent LinkedIn post offers a concise primer on context engineering—highlighting the fact that raw LLMs are "brilliant but amnesiac savants”: they possess extensive knowledge but retain no memory of you, your goals, or your previous work. The emerging discipline of context engineering is changing that by giving AI three critical capabilities:

• Memory—retrieving relevant information from past interactions

• Tools—connecting to databases and external systems

• Prompts—which have evolved into a form of natural-language programming

Seale argues that whether a fact lives in memory, a database, or a prompt, it only becomes meaningful when connected. One established approach for creating meaningful connections is the Knowledge Graph—a technology that has been around for years under names such as the Semantic Web and Linked Data. Connected relationships like these enable AI to become systems that genuinely understand your work environment.

If you’re just getting started with agents, this post is a perfect introduction.

What turns a large language model into an agent?
⚡ Quick Read (2 minutes)

AI Agents

How one content designer built a reliable AI agent

From Russel Norris at Intercom:

"I knew I wanted to use AI to scale myself without devaluing my or my team’s critical thinking, or flooding our product with generic, verbose content. I also knew building a good AI tool would do more than spit out individual pieces of microcopy or button text, and would need to be plugged into a system we could keep shaping."

Intercom's Russel Norris was facing a familiar problem: endless requests for content reviews, rewrites, and feature naming with no way to scale. Rather than treating AI as a one-off copywriting tool, he built VERBI—an AI agent integrated directly into their design system.

In his post, Russell provides a practical blueprint for building custom AI agents grounded in existing systems rather than generic LLMs. He explains how he set up VERBI to draw exclusively from curated sources, including style guides, glossaries, research, and, most critically, their Storybook design system, which features detailed content guidelines for every UI component. Using Glean's platform, he connected the agent to live documentation sources, ensuring it stays current without manual updates. The result: VERBI has been used over 700 times in just a few months.

I love his conclusion:

"The knee-jerk criticism of AI-driven content design is based on the assumption that good product teams just generate content from nothing and ship it. But great AI is the result of great human decisions. Its real value is to bring us together faster: getting us to a complete design we can consider as a team, before sharing it with the world. That’s how AI helps us win."

How to build a content design agent
☕ Medium Read (7 minutes)

That’s it for this week.

Thanks for reading, and see you next Wednesday with more curated AI/UX news and insights. 👋

All the best, Heidi