Recently, Anthropic published a post urging the world to consider slowing frontier AI development. The reason they gave was that Claude now writes more than 80% of its own code, compounding faster than the humans responsible for it can keep up with.
The pattern I see is that digital marketing programs get faster and faster at doing the thing, and slower and slower at understanding why it's working. The spend scales and nobody slows down to ask what's actually driving it. And somewhere in there, the person accountable for the program stopped being fluent in it. They're approving budgets based on a model they couldn't explain to you if you asked.
Anthropic called their version of this recursive self-improvement. The system gets good enough that it starts building its own next version without enough human judgment in the loop.
I don't think that's just an AI problem. This happens in marketing all the time.
At some point, it's worth asking: when did you last understand what's driving your program well enough to explain it to someone who can't see the dashboards?
My advice is to self-impose the friction before something or somebody else does it for you. Slow the cadence down, ask the uncomfortable question about what the number was actually measuring, and stay in the work instead of above it.
Winning the race isn't worth much if you've lost the ability to steer.
Patrick Patterson
CEO at Level Agency
What's interesting this week
Google's May 2026 core update raises the quality bar again Google's May 2026 core update wrapped up about three weeks ago, right on the heels of I/O. That timing isn't a coincidence. AI Overviews and AI Mode pull from the same index SEO has always optimized for. Content that doesn't earn visibility in traditional search doesn't show up in AI-generated answers either. March's update hit high-volume, thin-differentiation content programs hardest. This one is likely to do the same.
Key takeaway: Google has been saying the same thing for years. Original point of view, genuine expertise, a reason to exist that a generic AI summary can't replace. The update isn't new information. The cost of ignoring it is just higher now.
Level at SMX Advanced: what three days in Boston surfaced
We were in Boston last week for SMX Advanced. Two sessions, a booth activation built around a museum display case where attendees submitted what they thought would be the last keyword ever searched on a traditional search engine, and three days of practitioners wrestling with the same underlying tension: platform automation is outpacing the governance frameworks most teams have in place to manage it.
Key takeaway: The room was past debating whether AI changes search. The conversation has moved to what you can still govern, and what good inputs look like when the platform is making decisions you used to make yourself.
Most landing pages fail before the headline PPC Hero on a problem that shows up in almost every audit. Someone searches for pricing, clicks an ad, and lands on a page built for someone still comparing options. The friction is invisible in the platform and shows up in the conversion rate. The piece makes the case for mapping landing pages to search intent stages rather than campaigns or products, because the same product means something completely different depending on where a buyer is in the decision.
Key takeaway: Before you test the headline, check whether the page was built for the traffic you're sending to it. Most landing page problems aren't design problems.
The customer signal problem nobody has actually solved
CMSWire published a piece this week on something I keep running into. The concept is a "contextual intelligence loop": connecting your CRM, digital experience platform, and support tools so customer signals automatically feed back into the experience, instead of dying in a handoff between departments.
When a marketer spots a problem in the post-click experience, the fix typically runs through three departments. Marketing flags it. Support owns one piece. Web ops owns another. Each handoff takes days. By the time the fix ships, the data behind it is two weeks old.
This is especially costly in any program where the post-click experience is doing real conversion work. Higher ed inquiry flows, financial services lead nurture, B2B demo paths. The faster you close the loop between signal and action, the less ground you give back after acquisition.
Key takeaway: The breakdown usually happens in the friction between flagging a signal and having the authority to act on it. Fixing the experience requires solving the ownership problem within that feedback loop.
What higher education marketers are optimizing for (and why it's the wrong thing)
We published a piece this week on higher education marketing, but the real audience is anyone running campaigns where the conversion takes longer than a click to prove its worth.
Google built its ad engine for e-commerce. Instant signal, instant feedback, instant optimization. Higher education doesn't work that way. A lead might take months to reveal whether it was worth anything. If you're optimizing toward a generic form fill, you're not telling the platform what you want. You're telling it what you settled for.
The social side is the same problem in a different costume. Creative is now the targeting. Four or five ads won't cut it. And measuring each channel in its own silo is how you defend last quarter's numbers without ever understanding what actually drove them.
Key takeaway: The platforms will optimize for whatever signal you give them. The question is whether that signal has anything to do with the outcome you actually care about.
Google updated NotebookLM yesterday. It now runs on Gemini 3.5, has a built-in cloud computer for writing and executing code, and can turn whatever you feed it into PDF reports, spreadsheets, slide decks, and charts. You can also start a project with a loose question and let it find and organize material from the web rather than coming in with everything pre-assembled.
For brand marketers, it cuts hours off research. The gap between having a bunch of documents and having something you can hand off closes considerably faster now.
Free to use. The new features are rolling out to Google AI Ultra and Workspace accounts first.
Your AI visibility, scored.
When did you last audit what AI thinks of your brand?
Don't rely on an unreviewed algorithm to manage your brand’s reputation. Our scorecard pulls back the curtain on how AI platforms frame your brand sentiment and which sources they use to vet your authority.
Inside your custom report:
Visibility score and share of voice: The percentage of AI prompts your brand appears in, benchmarked against your top competitors across all major models.
Sentiment and positioning: Whether AI describes your brand favorably, neutrally, or negatively, and whether you're showing up as the primary recommendation or an afterthought.
Optimization roadmap: The specific content structure and authority signals you need to improve your score.
