The AI agent your vendor just demo’d is probably a for-loop with a login token — and this week’s signals expose the gap between the pitch and production reality. From O’Reilly’s blunt audit of agent architecture to a practitioner hitting $1M ARR with Claude Code, the clearest picture yet of who’s actually winning with AI is coming into focus.
Most “AI Agents” Are Fragile Plumbing in Disguise
O’Reilly’s essay pulls back the curtain on production AI agents: underneath the marketing, you’ll find fragile session logic, shared service accounts, and a security model held together by hope — not intelligence. For any marketing operations team evaluating agentic vendor tools right now, this means reliability and security posture must be first-order criteria before a single workflow dependency is built on top.
Before approving any AI agent integration this quarter, demand an architecture review covering session management, service account permissions, and failure-state behavior — not just a polished demo.
Read the full story →
Join the discussion →
HubSpot Names the AI Search Attribution Gap
HubSpot formally identifies AI search analytics as the structural missing link between organic traffic reports and pipeline data — a new tool category that didn’t formally exist twelve months ago. As AI-generated answers absorb top-of-funnel clicks, GA4 and Search Console can no longer explain the full story, meaning teams flying blind here will misattribute demand and misallocate budget in ways that compound over time.
Audit your analytics stack this week to check whether you have any instrumentation for AI search visibility — if GA4 and Search Console are your only sources of truth, you have a growing blind spot.
Read the full story →
Join the discussion →
Claude Code + Obsidian Becomes a Full Personal OS
A practitioner has extended the Claude Code and Obsidian integration into a fully custom personal operating system — complete with a Claude-built plugin, integrated terminal, live audience analytics across platforms, Hacker News pulse tracking, and content opportunity mapping, all inside a single vault. The recursive signal here is that the custom Obsidian plugin was itself built by Claude Code, meaning the tool is actively improving its own operator’s workflow in real time.
If you use Obsidian for knowledge management, dedicate a focused session this week to exploring the Claude Code terminal integration — the documented workflow covers audience tracking, research aggregation, and content opportunity identification in one environment.
Read the full story →
Try it yourself →
One Practitioner Hit $1M ARR Building SaaS with Claude Code
Nick Saraev documents building Clarbo — an AI-enabled power dialer SaaS — to $1M ARR using Claude Code, and has published a complete mobile app development course covering iOS, Android, React Native, and App Store submission with no prior mobile experience required. This is a documented proof point, not a concept demo: Claude Code can carry a product from zero to production revenue, including full mobile deployment.
If you have a marketing tool, internal dashboard, or audience utility waiting on engineering resources, treat Claude Code as a viable first-build option — the documented success path now covers SaaS, mobile, and App Store deployment.
Read the full story →
Try it yourself →
AEO Is Now a Formal Discipline — Not Just a Buzzword
Search Engine Journal formalizes AEO — now specifically framed as Agentic Engine Optimization — as a distinct discipline with new content visibility rules for AI search engines in 2026. The “agentic” upgrade is meaningful: content must now be structured to be retrieved and acted upon by AI agents, not merely cited in a summary answer, which changes content architecture requirements at a structural level affecting schema markup, information density, and page-level clarity.
Review your highest-traffic content assets through the lens of agent retrievability — ask whether an AI agent could parse, act on, and confidently cite each piece, and flag anything optimized purely for human reading patterns.
Google Brings AI Shopping Visibility Into Merchant Center
Google launches AI Performance Insights and Conversational Attributes in Merchant Center, giving retailers their first native tool to measure product discovery and visibility across AI-powered shopping surfaces. This is Google officially instrumenting the AI shopping dark funnel — “did my product appear in an AI answer?” is now partially measurable inside Google’s own ecosystem, and performance marketers who ignore this new reporting layer will be flying blind as AI shopping surfaces expand.
If you manage or advise any e-commerce accounts using Google Merchant Center, activate and review the AI Performance Insights panel this week — early fluency with the new data model will be a competitive advantage as AI shopping attribution becomes a standard KPI.
Most Teams Still Misread GA4 — and AI Search Makes It Worse
Neil Patel’s GA4 expert reports guide surfaces a live competitive gap: most marketing teams still aren’t using the reports that connect traffic behavior to funnel outcomes, three years after the Universal Analytics sunset. As AI search erodes top-of-funnel traffic simultaneously, teams that can’t accurately read their own funnel data will compound the attribution problem — misreading GA4 in a declining organic environment produces compounding strategic errors, not just reporting noise.
Use this guide as a quick audit checklist to identify which expert-recommended GA4 reports your team isn’t regularly reviewing, and prioritize funnel analysis and user journey reports given the current traffic environment.
Read the full story →
Join the discussion →
Gemini 3.5 Flash Goes GA — and Karpathy Joins Anthropic
Gemini 3.5 Flash reaches generally available status with Google explicitly positioning it for frontier agentic execution and long-horizon tasks, removing the experimental-tier risk for enterprise teams. In the same news cycle, Andrej Karpathy joining Anthropic arrived as a significant talent signal — when someone who could work anywhere chooses Anthropic over Google or OpenAI, it suggests serious AI research momentum is consolidating there even as Google ships production infrastructure. Community benchmarking already flags that Gemini 3.5 Flash performs below its official numbers on third-party evaluation platforms, making independent testing before deployment non-negotiable.
Add Gemini 3.5 Flash to your model benchmark set this week for marketing automation or content pipeline use cases — but run your own third-party evaluation before committing it to any production workflow.
Read the full story →
Try it yourself →
Join the discussion →
OpenAI Model Breaks an 80-Year-Old Math Conjecture
An OpenAI model disproved the unit distance problem — an 80-year-old conjecture in discrete geometry — marking the first credibly documented case of AI producing genuinely novel mathematical knowledge, not just reasoning over known facts. For content creators and strategists, the near-term implication is that AI-generated research and analysis will become increasingly difficult to dismiss as derivative, and “original insight” as a content differentiator faces a new kind of competition within the next few years.
Watch how OpenAI documents the methodology behind this result — if they reveal which model capabilities drove the discovery, it will directly inform which models to select for research-intensive content and strategy workflows.
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.