BARNEY GOODMAN
Back to Blog
/4 min read

The Day Anthropic Put White-Collar Work on Notice

Technologydigital-transformationfintechtechnologyAI

Claude Cowork plugins are reshaping white-collar work. Here's what's actually happening and why it matters.

In late January, Anthropic released 11 open-source Claude Cowork plugins. Then in late February, they added a dozen more. If you work in finance, law, HR, design, or engineering, and you haven't paid attention yet, now's the time.

The market noticed immediately. Thomson Reuters shares dropped 18% the moment the legal plugin shipped. IBM fell 13.2% in a single day when Anthropic announced Claude Code could automate COBOL modernisation. ServiceNow down 23%, Salesforce down 22%, Intuit down 33%. These are trillion-dollar companies watching their traditional moats vanish.

I've been using Claude for two years now. I've built things that would have taken months in weeks. I've compressed decision timelines, automated entire workflows, and done work that previously needed hiring contractors. But I'm not representative. I chose to work with this technology. Most people didn't get a choice. That's the story worth telling.

The Contract Problem

The legal plugin matters because it automates contract review, NDA triage, compliance workflows, and templated legal responses. This sounds abstract until you realise what it means: junior associates who spend 80% of their time reviewing documents now compete with a tool that reviews them in seconds.

LexisNexis integrated the plugin into its Protégé genAI suite within weeks. The logic was defensive: if they didn't, a competitor would, and their customers would leave. RELX, the parent company, watched its share price drop 14% on the announcement. The market was repricing what legal services are actually worth.

The firms that will survive are the ones that use this tool to do better work faster, not the ones pretending it doesn't exist. But survival isn't the same as stability. A firm that previously hired five junior lawyers to handle contract review now needs one person managing a Claude instance. That maths is brutal and simple.

COBOL and the IBM Moment

On February 23, Anthropic announced that Claude Code could automate COBOL modernisation. This matters because 95% of ATM transactions globally still run on COBOL. Airlines use it. Banks use it. Governments use it. The reason is simple: it works, it's stable, and replacing it costs millions of pounds and takes years.

IBM's entire consulting business has been built around charging customers enormous fees to slowly, carefully migrate these systems to modern platforms. That's where the money is. That's where it's always been.

When IBM's share price fell 13.2% in a single day, their worst performance since October 2000, the market was saying something specific: we don't believe in that revenue stream anymore. IBM is down 25% for the month. The market has repriced the entire business model.

Claude Code can't solve every COBOL problem overnight. But it can map dependencies, automate documentation, identify risks, and incrementally migrate code to modern languages. It compresses timelines from years to months. For the CIOs and CFOs responsible for these transformations, that changes the economics entirely. For the consultants who charged £500 an hour to sit in war rooms moving code by hand, it's a different picture altogether.

The Productivity Paradox

Here's the bit where I sound like I'm cheerleading, but I'll try to be honest about what I'm seeing.

Companies deploying customised Claude instances report 20 to 40% productivity gains. Early Cowork adopters, particularly in IT services, report 40 to 60% gains. Marketing, operations, and HR teams report 20 to 30% time savings. Anthropic internally reports 50% productivity gains in their own development work.

These numbers are real. I've seen them in my own work. A task that took me three hours last year takes me 45 minutes now. I'm not smarter. The tool is better. So am I more productive, or is my job just disappearing?

The honest answer is both. I'm doing more strategic work because the tactical stuff is automated. But that only works if you're already strategic. If you're an entry-level analyst doing tactical work, you've become redundant.

The Warning Nobody Wants to Hear

Dario Amodei, Anthropic's CEO, has been unusually candid about what he thinks this means. He's said AI could eliminate 50% of entry-level white-collar jobs within five years. He's warned that unemployment could hit 20%. He's called it an "unusually painful" disruption. He's even proposed a 3% "token tax" on AI revenues specifically to fund redistribution of wealth to workers displaced by automation.

You don't hear CEOs talk like that when they're trying to sell you something. Amodei sounds like someone who is genuinely troubled by what he's built.

Deutsche Bank warned recently about "AI redundancy washing": companies using AI as a convenient narrative to justify layoffs that are actually driven by other economic pressures. The Yale Budget Lab examined US labour market data from 2022 to 2025 and found no significant shift in job distribution yet, despite ChatGPT's debut. So the disruption is coming, but it hasn't fully arrived. We're in the before picture.

The Honest Take

I'm not going to tell you this is wonderful news for everyone. It's not. I'm also not going to tell you we should stop building this technology. We won't, and trying to would be pointless.

What's actually happening is simple: tools that compress timelines and automate work are real. They create genuine productivity gains for people and organisations willing to use them strategically. They also eliminate jobs that consist primarily of executing tasks that can now be automated. Both things are true.

This technology is already changing work. The only question left is what we do about it.

Some industries are moving faster than others. Law is moving fast. COBOL modernisation is moving fast. Finance will move fast. The ones that don't move fast will watch their market value collapse, and the ones that do will discover they need fewer people to do the same amount of work.

This is already happening.

What's your industry doing about this? Are you using these tools, or are you waiting for someone else to decide what your job looks like?

Sources

2026