Why a coding tool is worth a trillion dollars

Anthropic filed for an IPO last Sunday.

The valuation: $965 billion — set by a $65 billion raise just days before the filing. Recent reporting puts run-rate revenue around $47 billion, and coverage suggests the listing itself could target a valuation near $1.8 trillion — potentially the largest IPO in history.

Most of the coverage has been about the funding rounds, the chip deals, the regulatory exposure. But if you want to understand where the growth actually came from, you just need to look at one product: Claude Code.

It's a terminal-based coding agent. You open a command line, describe what you want to build or fix in plain English, and Claude Code reads your project files, reasons through the problem, writes and edits code, runs tests, and fixes what breaks — without you having to know the syntax. Not a chat window with suggestions. An agent that does the work.

Developers are paying $100–200/month for it. And there are enough of them that it's driving the fastest revenue ramp in software history.

Here's why I'm writing about it: that's not the audience it was built for, and it's definitely not the only audience using it.

I've been running it for months. Not to ship production code at scale, but for things like building an automated trading system, scripting database jobs, creating services to deploy on Railway, wiring up automations that needed custom logic. Tasks where I had the idea and the domain knowledge but not the hours to write every line myself. Claude Code handles the parts I'd otherwise postpone, look up, or outsource.

The honest version: it's not magic. It works best on well-defined problems where you can clearly describe what you want. Complex debugging, architectural changes, navigating an unfamiliar codebase — that's where it shines. On ambiguous tasks, it can produce code that almost works and require several rounds of correction. You still need enough context to review what it ships.

But "enough to review" is a much lower bar than "enough to write from scratch." That gap is where the value is.

Who this is for: Anyone who has technical tasks they're currently outsourcing, avoiding, or attempting badly. If you've thought "I'd build this myself but I don't code" — Claude Code changes that math. The $20/month Pro plan runs Sonnet, which handles most tasks well. The $100–200/month Max plan gives you Opus, noticeably better on complex work.

If you're already paying for Claude, it's included with your plan — search "Claude Code" to get set up.

Anthropic filed for an IPO. Confidential S-1 submitted to the SEC on June 1 — a potential October 2026 listing. The valuation story is the headline, but the underlying business story is what matters: recent reporting puts run-rate revenue around $47 billion, driven heavily by developer tools and enterprise API usage. Claude Code was the growth engine. When analysts try to explain a $965B valuation, the answer is: companies are paying real money because it makes their developers faster. That's a different story than the AI hype cycle of 2023.

Microsoft Scout launched at Build 2026. An AI assistant that appears in your Outlook and calendar system as if it were a real employee — not a sidebar chatbot. Scout can respond to meeting requests, relay messages from colleagues, and act across Outlook, Teams, and SharePoint while you're not watching. It's rolling out through Microsoft's Frontier program — not broadly available yet — but this is the clearest public demo of what "AI coworker" looks like when it actually ships — as a participant on your calendar, not a tool you open.

[BOLD]Microsoft released its first internal reasoning model.[/BOLD] MAI-Thinking-1, built by Microsoft's AI Superintelligence Team, is their first in-house reasoning model — 35 billion active parameters, a 256K context window, and in blind tests independent raters preferred it to Claude Sonnet 4.6, while it matches Opus 4.6 on SWE-Bench Pro coding. Trained only on commercially licensed data; in private preview on Foundry now. The pattern to notice: every major tech company now has its own internal model team. "We'll just use OpenAI" is becoming "we'll use OpenAI as a baseline."

Make AI Agents expanded to all paid plans. Make — the no-code automation platform — shipped their agents feature broadly this year. You can build, run, and chain autonomous agents directly in their visual canvas without writing code. The key upgrade over standard automation: agents can make decisions mid-workflow rather than just executing fixed steps. If you run Zapier or n8n and have wanted agent behavior without a custom build, this is worth testing.

Before you build an automation — or pay someone else to build it — run this. Pick any task you do manually on a regular basis, describe it clearly, and paste:

This does two things: forces you to define the task clearly (which most people skip and then wonder why the automation doesn't work), and gives you a rough spec you can hand off, drop into Claude Code, or build yourself. I run a version of this before touching any new automation project.

The output usually makes one thing obvious: the "hard part" is almost never the automation itself. It's one ambiguous handoff in the middle that you've been handling manually without thinking about it.

Like what you're reading? Forward it to someone who'd get value from it. And if you're curious what AI automation could look like for your business, book a free 15-minute audit — no pitch, just a look at where you're leaving time on the table.