Anthropic's head of Claude Code just revealed that 90% of the tool's codebase is written by the tool itself. For marketing and commerce teams, this isn't a developer curiosity. It's a preview of what's coming for every AI-powered system in your business.
Boris Cherny, the head of Claude Code at Anthropic, dropped a bombshell on Lenny's Podcast: he hasn't written a single line of code by hand since November 2025. Every pull request - and he ships 10 to 30 of them daily - is authored entirely by Claude Code. Across Anthropic as a whole, between 70% and 90% of all code is now AI-generated, according to Fortune.
This isn't just a story about developers losing their jobs. It's a signal about where every AI-powered business tool is heading - including the ones managing your ad campaigns, optimizing your product listings, and determining whether AI assistants recommend your brand.
What "Self-Writing Code" Actually Means
Let's cut through the hype. Claude Code is an AI coding agent that lives in your terminal. You describe what you want in plain English, and it writes, tests, and ships the code. What makes the 90% figure remarkable is that the tool used to build Claude Code is Claude Code. The AI improved itself.
Andrej Karpathy - Tesla's former AI chief and an OpenAI co-founder - described a similar shift in his own workflow. In November 2025, he was writing 80% of code manually. By December, AI agents were handling 80% of the work. He said capabilities "crossed some kind of threshold of coherence" that triggered a fundamental change in how software gets built.
Dario Amodei, Anthropic's CEO, framed the current moment as a "centaur phase" - humans and AI working together, with AI handling the execution and humans providing direction. But he warned: "The period may be very brief." He predicts AI could handle "most, maybe all" of what software engineers do within six to twelve months.
Bloomberg called it "The Great Productivity Panic of 2026." But for brands and marketing teams, panic is the wrong response. Preparation is the right one.
Why Marketing Teams Should Pay Attention
Here's the connection most coverage misses: the same AI that writes its own code is already being used to build marketing automation tools, and those tools are getting better at an accelerating rate.
Anthropic's own growth marketing team is a case study. Austin Lau, a growth marketer at Anthropic who had never opened a terminal before, now uses Claude Code to process CSV files containing hundreds of Google Ads, identify underperformers, and generate new ad variations - all within strict character limits for headlines (30 characters) and descriptions (90 characters). According to Anthropic's marketing blog, the workflow reduced per-ad creation time from approximately 30 minutes to 30 seconds.
That's not a 2x improvement. That's a 60x improvement. And it was built by someone with zero coding experience, using a tool that's improving itself.
Now multiply that across every marketing function: SEO audits, content optimization, product feed management, bid automation, competitive monitoring. When the tools building these workflows can improve their own code, the compounding effect is unprecedented.
Three Areas Where Self-Improving AI Changes the Game
1. AI Visibility and SEO
Traditional SEO meant optimizing for Google's algorithm. In 2026, you're also optimizing for ChatGPT, Perplexity, Claude, and Gemini - AI engines that decide whether to recommend your brand when consumers ask questions. This is Generative Engine Optimization (GEO), and it's fundamentally different from ranking in ten blue links.
Self-improving AI tools change GEO in two ways. First, the AI agents doing the recommending are getting smarter and more selective about which brands they cite. According to Princeton's GEO research, content with verifiable statistics and authoritative citations is 30-40% more likely to be surfaced by generative engines. Second, the tools you use to monitor and optimize your AI visibility - like auditing your brand's presence across AI platforms, tracking citation patterns, and adjusting content structure - can now improve themselves over time.
For CPG brands and e-commerce companies, this means the gap between brands that invest in AI visibility now and those that wait will compound faster than anyone expected.
2. Paid Media and Advertising
If Anthropic's own marketing team is using Claude Code to generate and optimize ad creative 60 times faster, what happens when that workflow improves itself quarter over quarter? The major ad platforms are already proving the answer.
Amazon launched Performance+ for its DSP, and the results speak for themselves: a 51% improvement in customer acquisition costs compared to legacy Amazon DSP campaigns, according to Amazon's internal data. Brands using Amazon's AI-powered creative tools advertise five times more products and use twice as many images per product compared to those that don't. On the search side, Amazon opened its Ads MCP Server in open beta in February 2026, letting AI agents like Claude create Sponsored Products campaigns, pull performance reports, and expand across marketplaces through natural language commands. Early beta testers report 67% faster campaign launches.
Meta is on a similar trajectory with Advantage+. According to Meta's earnings data, Advantage+ campaigns deliver $4.52 return per $1 spent, 22% higher than manual campaigns. Meta's AI infrastructure improvements drove 21% growth in ad revenue in 2025, and the company plans to roll out fully automated AI-driven ad creation by the end of 2026, generating complete ad assets from a website URL, selecting audiences without demographic inputs, and allocating budget to top performers automatically.
Google is pushing the same direction with Performance Max. In Q1 2026, PMax campaigns managed over 80% of total ad spend for the median enterprise account, up from 55% in 2024. Google reports PMax generates 35% more conversions at 20% lower CPA compared to equivalent manual campaigns. Advertisers using Target ROAS Smart Bidding achieve 38% higher return on ad spend than those using manual CPC bidding, per cross-industry benchmarks.
The pattern across all three platforms is identical: AI handles more of the execution, performance improves, and the human role shifts from campaign management to strategic direction. This is the centaur phase for advertising. The brands that build proprietary AI workflows on top of these platforms - connecting Amazon, Meta, and Google data into unified optimization loops - are investing in assets that compound in capability. A custom cross-channel agent that improves its own decision-making logic isn't just a tool. It's a moat.
3. Agentic Commerce and the Digital Shelf
The convergence of self-improving AI tools and agentic commerce - where AI agents make purchasing decisions on behalf of consumers - creates a compounding effect that changes how products get discovered and bought.
When AI shopping agents evaluate products, they parse structured data, read reviews, compare specifications, and check availability - all programmatically. The tools that optimize your product listings, manage your pricing, and maintain your digital shelf presence are increasingly AI-powered. When those tools can improve their own algorithms, your digital shelf optimization compounds without proportional increases in human effort.
For consumer brands, especially in CPG and grocery where AI-driven purchasing is projected to grow fastest, building AI-powered commerce operations now creates an exponentially growing advantage.
The Slopacolypse Warning: Why Human Expertise Still Wins
Not everyone is celebrating. Karpathy predicted a 2026 "slopacolypse" - a flood of AI-generated content across GitHub, Substack, arXiv, and the broader internet that "generally tends to work but is often low quality." He warned about masses of "almost right, but not quite" output that will dramatically increase the cost of filtering signal from noise.
This is exactly why expert-guided AI outperforms fully autonomous AI. The brands that treat AI as a replacement for human judgment will produce commodity slop. The brands that pair AI execution speed with human strategic direction - the centaur model Amodei described - will produce work that stands above the noise.
In practical terms: an AI tool that generates 50 ad variations in 30 seconds is only valuable if a human marketer with domain expertise is setting the strategy, defining the brand voice, and evaluating the output. The "architect" role - defining what to build and why - becomes the competitive advantage, not the "builder" role of writing individual lines of code or copy.
What You Should Do Now
Audit your AI tool stack. Are you using static tools that require manual updates, or are you building with platforms that improve over time? The difference compounds dramatically over 12-24 months.
Start building proprietary AI workflows. Even simple automations - like using Claude Code to process ad performance data or generate SEO recommendations - create institutional knowledge that compounds. Anthropic's own growth team proved this works even for non-technical marketers.
Invest in AI enablement for your team. The "architect" role is the new competitive advantage. Train your team to direct AI tools effectively, not to compete with them on execution speed. According to Anthropic's enterprise research, organizations see 2-3x improvement in shipping speed when they treat AI as a team member rather than just a tool.
Monitor your AI visibility. As AI tools improve, so do the AI agents recommending products and services to consumers. If your brand isn't being cited by ChatGPT, Perplexity, and other AI engines today, it won't be part of the conversation tomorrow. Regular AI visibility audits are no longer optional.
Automate your paid media across platforms. Amazon, Meta, and Google are all pushing toward AI-driven campaign management. Brands that connect these platforms into unified, AI-powered optimization loops - automating bidding, creative testing, and budget allocation - build compounding advantages that manual campaign management can't match.
Don't wait for the tools to be perfect. Cherny himself acknowledges you "cannot be totally hands-off." The tools need human oversight. But the teams that start building with self-improving AI now will be miles ahead of those that wait for a fully polished solution.
The Bottom Line
Claude Code writing 90% of its own codebase isn't a curiosity - it's a leading indicator. The same self-improvement dynamic is coming to every AI-powered marketing tool, ad platform, and commerce system. The question isn't whether your tools will improve themselves. It's whether you'll be building with them when they do.
Texin.ai helps consumer brands build AI-powered marketing stacks that compound in capability - from AI visibility audits and GEO optimization to automated ad campaigns and agentic commerce readiness. Talk to us about turning self-improving AI into measurable growth.
Ready to put this into practice?
We manage advertising across Amazon, Meta, Google, and more for brands that want AI-powered growth.
