Markets panicked when Claude got plugins. I think they missed the point. The real story isn’t disruption theatre — it’s that Anthropic has built the first genuinely coherent AI workflow stack for serious knowledge work, and broadcast is barely paying attention.

By Ben Anchor — Wednesday, 4 March 2026 · Listen to the podcast episode

The market can smell fear, even when it can’t quite articulate the source. On 30th January, Anthropic added plugins to Claude Cowork, and $285bn evaporated from enterprise software stocks in a single session. Thomson Reuters, RELX, Wolters Kluwer — information and data companies, not software vendors in the traditional sense — all shed significant value. Traders dubbed it the SaaSpocalypse, which tells you the anxiety was already there, coiled and waiting for a trigger.

But I think the panic actually obscured the more interesting story, which is that Anthropic has quietly built the most coherent end-to-end AI architecture for knowledge work that exists right now, and most of broadcast is still treating Claude as a chatbot.

Raiyana and I unpacked this in the latest episode of Broadcast Media: The Inside Track, and what struck me afterwards wasn’t the market drama — that’s noise — but how few people in our industry seem to understand what Anthropic is actually building, or why it matters for the kind of work we do.

Three Products, Not One

Let’s start with the architecture, because this is where most people’s mental model breaks. Anthropic now ships three distinct products, each operating at a different level of autonomy:

Claude AI is the thinking layer — the web interface, the app, the conversational entry point. This is where most users start, and for many, it’s where they stay. It’s good at that: long-form reasoning, document analysis, editorial structure. I use it daily for podcast ideation, script shaping, and working through dense research material. The long context window is genuinely useful when you’re trying to extract signal from a 40-page policy document or a sprawling interview transcript.

Claude Code is the developer tool — terminal-based, agentic, built for technical workflows. It doesn’t just generate code; it executes it. It reads your file system, writes scripts, runs them, iterates on output. For anyone managing workflow automation (I use it for N8n integrations), this is where the tool earns its place. The recent update added remote control — you can hand off a live terminal session to your phone or browser and resume from any device. That’s not a feature; it’s a shift in how you actually use the thing.

Claude Cowork is the newest layer, and the one that spooked the markets. It’s an agentic desktop application designed for non-technical knowledge workers. It runs tasks in the background, across your files and connected tools, with real autonomy. The shorthand Anthropic uses is clean: use Claude for thinking, Code when the work lives in a repository, Cowork for cross-application workflows.

Three products. Three operating modes. Progressively more autonomous. That’s a stack, not a feature set.

Why the Market Panicked (and Why It Was Wrong)

The SaaSpocalypse reaction makes sense if you squint. Anthropic’s customer base is heavily concentrated in exactly the sectors that got hammered: legal research, financial data, regulatory intelligence. Thomson Reuters has been integrating Claude into their legal product Co-Counsel for contract analysis and drafting. So you have the parent company’s stock falling partly because their own product division is already building on the technology perceived as the threat. That’s a complicated position.

But the panic was premature. Cowork in its current form has real friction: macOS only, no cross-device sync, plugin configuration requires editing markdown files and wiring up integrations manually, and the token budget burns fast. The $100/month max plan is the realistic price point for serious use. Most knowledge workers aren’t going to self-configure this today.

The analyst view that’s settled in since is that it was an overreaction to a research preview. And I think that’s right. The pattern we keep seeing with AI product launches is: dramatic announcement, over-corrected fear, messy reality check, then gradual deep integration that actually changes things. We’re somewhere between the second and third phase on Cowork. The technology is capable, but not frictionless. The transformation is coming, but it’s not instantaneous.

And that, frankly, makes this a useful moment for broadcast organisations. The window to build competency hasn’t closed. The companies investing in understanding the Anthropic stack now — deploying Claude in editorial workflows, putting Claude Code in technical teams, watching how the plugin architecture develops — those are the ones positioned to move quickly when the friction reduces.

What Anthropic Actually Sells (and to Whom)

It’s worth stepping back and asking who Anthropic is building for, because the answer matters. The enterprise business is substantial, and it’s concentrated in sectors that require serious analytical rigour: government and defence contractors, healthcare systems, financial services, legal tech. These aren’t consumer-facing products. They’re mission-critical systems where reliability and governance aren’t nice-to-haves.

The reason those sectors chose Anthropic over alternatives often comes back to the safety-first positioning. Constitutional AI — training models to reason within defined principles rather than purely optimising for human approval — is the research framework underpinning everything they ship. Claude tends to feel more measured, more considered. There’s structure underneath the responses, not just pattern matching.

And then there’s the ad-free commitment. Anthropic ran a Super Bowl ad in February — roughly $8 million for the slot — with a single message: Claude will not run ads. While Meta, Amazon, and Google were all buying spots in the same break, Anthropic used the platform to make one promise about what Claude is not. For enterprise customers in regulated industries, that promise has real commercial value. If you’re a broadcaster deploying AI in editorial or compliance workflows, you want confidence that the model’s outputs aren’t being shaped by commercial interests you can’t see. The ad-free commitment is a governance signal as much as a marketing one.

Where Broadcast Fits (and Where We’re Missing the Boat)

I’ll be direct: most broadcast organisations I talk to are still treating AI as a novelty or a threat, rather than a set of tools you deploy to extend capability. The use cases are real and growing: research, document analysis, workflow automation, script development, compliance review. The entry points are low-friction. The ceiling is high.

For anyone wanting to build that internal capability, Anthropic launched a full course catalogue through Skilljar this month: free courses covering Claude fundamentals, Claude Code, the API, MCP integration, agent skills. That’s structured learning directly from Anthropic, without significant budget. Lowering the barrier to competence means faster adoption, which means more enterprise customers who understand what they’re buying. It’s good product strategy as much as goodwill.

The organisations building that competency now — embedding Claude in editorial teams, deploying Code in technical workflows, experimenting with the orchestration layer as it matures — those are the ones that will have a genuine advantage over those waiting for it all to become simpler. Because it won’t get simpler. It’ll get more capable, more autonomous, and more integrated. The question is whether you’re building fluency now or playing catch-up later.

The SaaSpocalypse wasn’t the story. The story is that Anthropic has built a coherent, safety-first AI workflow architecture for serious knowledge work, and broadcast is barely paying attention. That won’t last. The question is what you do with the window while it’s still open.