It's already happening...
A simple markdown file destroyed $285 billion in three days. Everything I wrote in Issue #1 is already happening.
On February 10, I came across a video that confirmed everything I wrote in Issue #1.
It was about how a simple markdown file from Anthropic destroyed $285 billion in market value in just a few days. The video was by Nate B Jones, and he explained what happened better than anything I’d read.
I've been reading about this shift for a while now. Industry leaders talking about AI agents replacing traditional software, articles about the agentic web, product announcements hinting at what's coming. In my first issue, I wrote about what I was learning: the web moving from "humans using software" to "humans sending agents."
But reading about something and seeing it actually happen are two different things.
The $285 billion crash? That was real. That was proof.
And honestly, it helped me understand what all those articles and announcements were actually pointing to.
I spent the rest of the week trying to connect what I was seeing. I watched that video multiple times. I read through product releases from OpenAI, Anthropic, and Google.
Let me share what I’m learning.
The $285 billion crash - what actually happened
On January 30, 2026, Anthropic released a few plugins for Claude Co-work.
Three days later, $285 billion in market value was gone.
Thomson Reuters, LegalZoom, Wolters Kluwer - all their shares dropped massively. Indian IT companies like Infosys and TCS also crashed hard.
I was confused at first. Why would a simple plugin release cause this?
Then I watched that video breakdown, and it clicked.
The plugin wasn’t the problem. The plugin exposed the problem.
One of those plugins was for legal work. It automates contract review, NDA triage, compliance checks. But here’s what shocked me: it’s just a 200-line markdown file. No fancy proprietary model. Just structured prompts telling Claude how to behave like a paralegal.
The video explained something I hadn’t fully understood: enterprise software is built on per-seat licensing. Every human who touches the tool pays a license fee. Thomson Reuters charges per user. Salesforce charges per user. Almost every SaaS tool charges per user.
That model worked when humans were the main users of the software.
But now, AI Agents are the users…
AI agents don’t need seats. One agent can do the work of 10 people, 50 people, 100 people. And it only needs one login.
If one AI agent replaces 10 paralegals, that’s 9 fewer Software licenses. That’s 9 fewer seats to charge for. Multiply that across every enterprise, every industry, every SaaS tool built on per-seat pricing.
The entire business model breaks.
This is what I was reading about in all those articles. But watching $285 billion disappear in three days made it real.
What I found while reading product releases from OpenAI, Anthropic, and Google
I was reading through announcements from the big AI companies. And I realized something: they're all building the same thing from different angles.
Let me show you what I mean.
OpenAI Frontier
I read through OpenAI’s announcement for Frontier. It’s not another chatbot product. It’s a platform for managing AI agents across your entire company.
What caught my attention:
It connects all your business systems (CRM, data warehouses, ticketing tools)
It treats agents like employees - with onboarding, permissions, shared business context
Companies like Oracle, Intuit, HP, and Uber are already using it
OpenAI also shared some outcomes of the companies that are already using it; a financial services firm reported getting 90% of their time back for client-facing teams. Another tech company saved 1,500 hours per month.
My connection to agentic web: This is the “operating system” layer for agents. Just like humans need operating systems, companies will need platforms like this to run agent teams.
Claude Opus 4.6 & GPT-5.3-Codex
OpenAI and Anthropic released competing models within an hour of each other.
Claude Opus 4.6 introduces “agent teams” - multiple agents working together on big tasks. GPT-5.3-Codex is 25% faster and handles long-running workflows.
But then I read something in the IBM Think Newsletter that reframed everything:
“A lot of the recent progress has not been on the models themselves. The real progress has been on the tooling, on the prompts, on the agents, on the MCP servers.”
I got it.. The race isn’t about smarter models anymore. It’s about better infrastructure for agents.
Google Chrome with Gemini 3 Auto Browse
Then I saw Google’s Chrome update.
They’ve integrated powerful new AI features in Chrome. One of them is “auto browse” (available only for US users). And when I read through what it does, I realized: this is the shift happening right in front of us.
Auto browse lets you send an AI agent to handle multi-step tasks in your browser while you do something else:
Scheduling appointments
Finding the right products and ordering for you
Filling out forms
Getting quotes from plumbers and electricians
Managing subscriptions, etc.
One of their example that hit me:
Let’s say you’re planning a Y2K theme party. You show it a photo. Auto browse identifies what’s in the picture, searches for similar items, adds them to your cart, stays within budget, and even applies discount codes.
This isn’t enterprise software. This is regular people sending agents to browse for them.
Remember in Issue #1, I wrote about the web shifting from “humans using browsers” to “humans sending agents”?
This is it. It’s happening. Not in some future, now.
Google even created something called Universal Commerce Protocol (UCP) - an open standard they co-developed with Shopify, Etsy, Wayfair, and Target so agents can take actions on your behalf seamlessly.
What this made me realize: The infrastructure isn’t just for enterprises. It’s for everyone. Google is betting that millions of people will stop clicking through websites and start sending agents instead.
The pattern I’m seeing
After watching that video and reading all these announcements, I started seeing a pattern.
What happened to print media is happening to enterprise software.
Print media didn’t die because content became worthless. The New York Times still has great journalism. What died was the access model - paying for physical newspapers.
The same thing is happening now:
Thomson Reuters’ legal databases are still valuable
Salesforce’s customer data is still valuable
The data isn’t going away
What’s dying is the per-seat access model for that data.
The video mentioned KPMG using AI as leverage to negotiate a discount on audit fees. The message: “You can do more with fewer people now. Adjust your pricing or we’ll find someone who will.”
And it’s not just enterprises. Chrome Auto Browse means regular people like you and me will also stop paying for many services. Why pay for a travel booking service when your browser agent can compare flights and hotels for you? Why pay for a form-filling Virtual Assistant when your agent can handle it?
This is the beginning, not the end.
What this means for the agentic web
Here’s what I’m connecting:
For businesses:
The per-seat SaaS model is breaking
Companies are building agent infrastructure right now (Frontier, Co-work)
Data still matters, but access models are changing
Even consumer-facing businesses need to think about agents (Chrome Auto Browse, UCP)
For individuals: The same pressure applies to knowledge workers. If an agent can do 80% of your job, what's the 20% only you can do?
I’m asking myself these questions every week now:
What tasks am I doing that an agent could handle?
What skills do I have that agents can’t replicate?
How should we prepare our websites when agents do most of the browsing?
How does discovery work in the agentic web?
If you’re exploring this too, I’d love to hear what you’re learning. Feel free to share your thoughts.
- Farhad