AI agents don’t need more hype. They need a map

AI is coming for us, they keep saying. We’ll all have AI agents acting on our behalf, doing everything from our weekly grocery shopping to booking airline tickets. It’s gonna change everything, just like the web did!

So, why do AI agents still seem to suck at everything? I’ve written before about the trials and tribulations of getting an AI agent to book a dinner reservation. As of now, they’re so slow and clunky, it’s way easier to just fire up OpenTable and do it yourself.

To understand why AI agents still struggle getting stuff done, just step in a time machine and dial up the year 1995. Back then, the nascent “World Wide Web” was very much like today’s so-called “agentic” web – in short, a mess.

Welcome to another edition of Prompt Mode, your weekly AI newsletter.

I’m your host, Ben Patterson. Each week on Prompt Mode, I’ll be serving up analysis of the AI trends that matter to everyday users like you and me. Stay tuned for practical AI tips, hands-on experiences with the latest AI tools, and–you guessed it–prompts to help you get the most out of your AI assistants.

Thanks for reading, and if you like what you see, just sign up right here.

There were web sites popping up left and right, but no easy way to find them. Google was still years away, and staffers at Yahoo were building out web directories by hand. Web crawlers were starting to appear, but there weren’t any agreed-upon standards for web sites to tell those crawlers what they were all about. If you stumbled upon a site, you might not know what it was, who it was for, if it was credible, or whether it was even safe.

That chaotic, 1995-era state of the web is similar to where we are now with the “agentic” web. Just like users struggled to find URLs back in those days, so do AI agents struggle to find the tools they need to actually get things done online. When you ask an AI agent to book you a dinner reservation and it can’t, that’s a symptom of an agent that can find the right tool for the job.

That brings us to a new standard called Agentic Resource Discovery, or ARD for short. Developed with the backing of Google, Microsoft, Nvidia, and other tech giants, ARD is a standardized “discoverability” layer that allows an online service to say to the world, “Here’s what I am, here’s what I do, and here’s my menu of AI tools.” That information could then be crawled, indexed, and ranked by agentic discovery services, which in turn could be searched by AI agents.

So yes, we’re essentially talking Google Search for AI, which would make it way easier for an AI agent to — among many other things — find a reputable and agent-ready restaurant reservation service, discover which tools it offers, employ the right authentication methods, and book that table, all on its own.

Can ARD help tame today’s chaotic agentic web? Google, Microsoft, and Nvidia sure hope so. But if ARD isn’t the answer, something else needs to be.

Prompt of the week: The “compress this chat” prompt

Whether you’re using ChatGPT, Claude, or Genini, your favorite AI chatbot will start to forget things if your thread gets too long. With their limited context windows, today’s LLMs can only remember so much at once, and once a conversation grows to a certain size, something has to go.

Luckily, the best AI tools (like the ChatGPT and Claude clients) can automatically compress lengthy chats. But there’s also a way to make any AI chatbot “compact” a thread that’s grown too long, allowing you to start fresh in a new chat that “reboots” the thread while retaining the essentials of your conversation. All you need is this prompt.