This week, two stories that seem unrelated — Anthropic’s AI writing 80% of its own code and Meta’s AI agent handing over Instagram accounts to hackers — are actually the same story. AI agents are gaining write-access to consequential systems faster than security, regulation, or most marketing teams can respond.
Claude Writes 80% of Anthropic’s Code — For Real
Anthropic’s “When AI Builds Itself” report confirms that over 80% of its production code is now written by Claude — not a benchmark, not a demo, but a live operational metric inside a top-tier AI lab. Simultaneously, the open-source tool Graphify, which converts codebases into traversable knowledge graphs, hit nearly 60,000 GitHub stars by solving Claude Code’s core memory and token-cost problem before Anthropic’s own roadmap addressed it.
Test Graphify this week if your team uses Claude Code for any marketing automation or internal tooling — claimed token savings of up to 70x could make complex projects economically viable that weren’t before.
Read the full story →
Try it yourself →
Meta’s AI Agent Handed Hackers Instagram Accounts
Attackers hijacked Instagram accounts — including the dormant Obama White House account — by simply asking Meta’s AI customer support agent in plain natural language to re-link those accounts to attacker-controlled emails, and it complied without any identity verification. No technical skills required: just a polite request to a compliant AI agent with write-access.
Audit every AI agent your team has deployed this week for write-permissions, and add human-in-the-loop confirmation gates for any action involving account ownership or access credential changes.
Enterprise AI Is in 1991 — The Window Is Open Now
Fast Company argues enterprise AI today mirrors the internet in 1991: the infrastructure is real and capable, but the equivalent of the Web browser that made it universally deployable has not arrived. Practitioners building AI-native workflows now are doing what early web-native businesses did before the browser — accumulating institutional knowledge that compounds in value the moment mass adoption arrives.
Treat your current AI workflow experiments as long-horizon infrastructure investment, not quarterly productivity wins — the practitioner who has internalized AI-native processes before the “browser moment” will have a structural career advantage that latecomers cannot quickly close.
One Week, Four Major AI Releases — Stop Evaluating Features
NVIDIA Nemotron 3 Ultra, Google Gemma 4 12B, OpenAI ChatGPT memory upgrades, and Cognition’s $10 million performance-backed guarantee for the Devin autonomous coding agent all shipped in a single week alongside Anthropic’s self-improvement claims. When this volume of major releases arrives simultaneously, individual feature evaluation is obsolete as a decision framework.
Map which of your core marketing workflows are now dependent on a specific vendor ecosystem — that dependency map is your actual strategic risk register, not a feature comparison spreadsheet.
Google GA4 Now Tracks GBP Calls and Direction Requests
Google formally documented a native link between Google Business Profile and Google Analytics 4, bringing local engagement signals — calls, direction requests — into GA4 reports for the first time since GA4 launched. This closes a measurement gap that previously required third-party call tracking workarounds and eliminates the silo between local organic presence and paid campaign attribution.
Connect your Google Business Profiles to GA4 this week and establish a baseline before your next local campaign launches — you now have before/after measurement capability that didn’t exist last month.
Read the full story →
Join the discussion →
Ahrefs Ships Nine Vibe-Coded Apps — Non-Developers Should Notice
Ahrefs published nine production-ready AI-built apps created by their own team using vibe coding, framing them as tools non-developers can install today to grow websites. When a major SEO platform’s own team is shipping internal marketing infrastructure via AI agent-assisted coding and publishing the results, the implicit message is that the “buy a vendor tool” vs. “build your own” calculus has fundamentally shifted.
Review Ahrefs’ nine vibe coding examples and identify which one maps to a real workflow bottleneck in your SEO or content process — the installation barrier is explicitly framed as low for non-developers.
Google Demand Gen Targeting Restrictions Are Silent Campaign Killers
Google issued updated guidance clarifying when sensitive audience targeting restrictions affect Demand Gen campaign delivery — the campaign type spanning YouTube, Discover, and Gmail that Google has aggressively promoted as the future of upper-funnel advertising. Silent delivery restrictions here don’t surface in standard campaign diagnostics, meaning practitioners may be attributing underperformance to budget or creative when the real cause is an undisclosed audience restriction.
Review all active Demand Gen campaigns against Google’s updated sensitive audience guidance this week, especially audience lists and interest signals that might be triggering delivery restrictions and explaining recent performance drops.
OpenAI Launches Lockdown Mode — Security Is Now a Product Category
OpenAI documented a new Lockdown Mode security feature in its help center, signaling that account-level security controls are now a product requirement driven by enterprise and high-risk user pressure — not an edge-case support ticket. The timing, in the same week Meta’s AI agent was exploited via natural language, is almost certainly not coincidental and reflects industry-wide recognition that agent-level access control is a front-line problem today.
Flag OpenAI’s Lockdown Mode to your IT and security leads now — enterprise AI procurement conversations will increasingly require security feature documentation, and knowing this exists before your IT team asks builds credibility.
A Psychologist’s 30-Year Warning About AI and Your Brain
Gloria Mark, a UC Irvine psychologist with three decades of empirical research on human-computer interaction, raised concerns at SXSW London that AI chatbots are causing people to lose control of their own cognitive processes and decision-making. This is not a think-piece — it comes from a researcher whose prior work on digital attention has held up empirically, and it is surfacing at a conference that shapes practitioner culture.
This week, audit which decisions in your marketing workflow you have delegated to AI outputs without active critical review — the cognitive atrophy risk Gloria Mark identifies is cumulative and quiet, not dramatic and sudden.
More from Rafal Reyzer
For deeper dives on AI and marketing strategy, visit my YouTube channel →
Rafal Reyzer
Hey there, welcome to my blog! I'm a full-time entrepreneur building two companies, a digital marketer, and a content creator with 10+ years of experience. I started RafalReyzer.com to provide you with great tools and strategies you can use to become a proficient digital marketer and achieve freedom through online creativity. My site is a one-stop shop for digital marketers, and content enthusiasts who want to be independent, earn more money, and create beautiful things. Explore my journey here, and don't forget to get in touch if you need help with digital marketing.