Here’s a number that got the tech world talking in early June 2026: according to Anthropic, the company behind the Claude assistant, its own AI model now writes more than 80% of the code actually merged into its systems. In other words, the AI that helps millions of people code now writes most of the software that runs it.

What makes this newsworthy isn’t just the percentage. It’s that Anthropic published it in a post warning that AI could soon start improving itself, and asking that we preserve the ability to pause development. When a company valued at hundreds of billions of dollars sounds its own alarm, it’s worth understanding why, and what it actually means for a small or medium business in Quebec.

1. Where the 80% figure comes from

The number comes from an Anthropic piece titled “When AI Builds Itself,” picked up by several credible outlets including Scientific American and Tom’s Hardware. The company explains that the share of code written by Claude and merged into its own systems went from a few percent, before the Claude Code tool launched in early 2025, to more than 80% today.

Anthropic adds another telling data point: its engineering teams reportedly ship about eight times more code per quarter than they did a few years ago. The human role hasn’t disappeared, but its nature has changed. Humans have shifted from writing every line to deciding, guiding, reviewing, and approving what the machine proposes.

• AI-generated code is still reviewed and approved before it gets merged.
• The productivity gain is real, but it comes with new oversight responsibilities.
• Anthropic itself notes that “we’re not there yet” and that full self-improvement “is not inevitable.”

2. What is “recursive self-improvement”

The term sounds like it’s straight out of a science fiction movie, but the idea is fairly concrete. Recursive self-improvement is the point where an AI becomes good enough to design and build an improved version of itself, with less and less human involvement. Each generation helps build the next one, faster.

Anthropic isn’t saying this has happened. The company says the trend points in that direction and that it would be reckless not to plan for it. That’s why it’s calling for keeping “the option to slow down or temporarily pause” development of the most advanced models, giving safety research and societal structures time to catch up.

Why would a company ask for a brake on its own product

This is the most surprising part. Anthropic has built its reputation on AI safety, and this post fits that pattern. Asking for a pause mechanism isn’t about distrusting the technology, it’s about recognizing that a technology accelerating this fast needs guardrails built in advance, not bolted on after the fact.

3. What this means for a Quebec small business

You might think this debate plays out far from the daily reality of a business in Trois-Rivieres or the Mauricie region. In reality, it directly affects how you’ll work with AI in the coming years.

• AI is becoming a real production collaborator. If it’s already writing 80% of the code at the company that built it, it can just as easily draft your emails, contracts, meeting summaries, or first drafts of documents.
• Human oversight still matters. The more AI handles, the more review and validation work matters. No one should publish or send AI-generated content without checking it first.
• Security can’t be an afterthought. Anthropic’s instinct, planning guardrails before you need them, applies to your business too: AI usage policies, access management, protecting sensitive data.

Adopting these tools without a framework is a bit like hiring a very fast employee and giving them no rules to follow. That’s exactly where structured IT support makes a difference. Our managed IT services help small businesses put AI to work in a useful and secure way, without exposing their data.

4. A milestone to put in perspective, not to fear

It’s worth keeping a level head here. The 80% figure applies to a very specific context: Anthropic’s internal code, written with its own models and its own review processes. It doesn’t mean all the world’s programming is happening on its own, or that developers have become obsolete. Human work has shifted toward design, direction, and quality control.

For a small business, the right takeaway is neither panic nor indifference. It’s informed curiosity. Companies that learn to use these tools methodically now will gain a real edge, as long as they keep a responsible person in the loop at every important step.

• Start with low-risk tasks: drafts, summaries, research.
• Clearly define what can, and can’t, go through AI in your organization.
• Train your team to spot a questionable answer and correct it.

Frequently asked questions

Does Claude really write its own code all by itself?

Not entirely on its own. According to Anthropic, Claude writes more than 80% of the code merged into its systems, but that code is still reviewed and approved by human engineers before going into production. Humans set the direction, and the AI carries out a large share of the work.

What is recursive self-improvement?

It’s the idea that an AI could design and build an improved version of itself, with less and less human help. Anthropic says we’re not there yet, but that the trend justifies planning guardrails now, including the option to pause development.

Should my business be worried about using AI?

No, as long as it’s managed properly. The real risk for a small business isn’t AI itself, it’s using it without rules: sensitive data shared by mistake, unverified content, poorly controlled access. With a clear policy and solid IT support, AI becomes a safe asset.

Adopting AI with a solid strategy

Anthropic’s announcement makes one thing clear: artificial intelligence is moving fast, and even its creators are asking for caution. For a small business, the winning move is to move forward with a plan, not on the fly. If you want to bring tools like Claude or Copilot into your operations while protecting your data, our team can build you a roadmap that fits. Discover our IT services for small businesses or reach out directly through our contact page to talk it over.