HAPPY FRIDAY,
Here's the breakdown on the latest shenanigans...
In This Week's AI Fridays:
- BCG's chief AI ethics officer says companies deploying agents have no idea what they're doing, and she's not wrong
- Wikipedia lost 5% of its search traffic to AI Overviews, and what that means for every content site you run
- ChatGPT gets a data agent, Slack gets a report builder, and the "vibe coding" trend has officially reached your Tuesday standup
- How someone analyzed hundreds of LinkedIn comments in minutes using AI browser automation
- Plus: Nvidia's projecting 70% growth, the UK economy credited AI for a good July, and a coding startup just hit a $48B valuation
Companies deploying AI agents have 'no idea how to manage risk,' AI safety expert warns
BCG's chief AI ethics officer went on record this week to say what a lot of people in this space have been thinking: businesses are racing to deploy agentic AI without any meaningful framework for understanding what could go wrong.
The Highlights:
- BCG's chief AI ethics officer calls out the gap between deployment speed and risk management capability
- Agentic AI is the focus: systems that take actions autonomously, not just generate text on request
- The concern isn't theoretical; no one has defined who is accountable when an agent does something wrong
- The warning comes from inside the consulting world that is actively selling AI transformation projects to those same companies
The Takeaway:
The gap between "we have agents running" and "we know what our agents are doing" is where the expensive mistakes live.
What Wikipedia Reveals About AI Overviews And Web Traffic
University of Washington researchers estimated that Google's AI Overviews have cut search referrals to Wikipedia by around 5%. Google disputes the methodology. Both things can be true, which is where we are. The reason Wikipedia matters here isn't sentiment; it's that Wikipedia is one of the most-linked, most-authoritative sources on the web, and if AI Overviews are taking traffic from them, the results for everyone else is probably worse.
The Highlights:
- University of Washington study estimates a ~5% drop in Google search referrals to Wikipedia attributable to AI Overviews
- Google disputes the finding, as Google tends to do when the finding is inconvenient
- Wikipedia is a useful proxy because its traffic is well-documented and its SEO authority is about as strong as it gets
- The implied question for marketers: if it's happening to Wikipedia, what is it doing to your blog, your resource library, your organic acquisition funnel
The Takeaway:
A 5% drop sounds manageable until you remember Wikipedia is the floor, not the ceiling. The sites with thinner authority are almost certainly seeing more.
AI in Marketing
Now everyone can put data to work
OpenAI rolled out a Data agent for ChatGPT Work this week, letting users connect their company data sources and build interactive dashboards through natural language. No SQL, no BI team ticket, no waiting three days for a chart that answers a question you had on Monday. For marketers, the specific appeal is the ability to pull your own data into the same environment where you're already doing analysis and writing, rather than toggling between five tools and copying numbers into a deck by hand. Whether it handles messy, inconsistent real-world data as smoothly as the demo suggests is the part that tends to reveal itself after the launch announcement. Worth getting hands on it now rather than waiting for the review cycle.
Slack can now vibe-code interactive charts and reports inside chats
Slack's new Slackforce Surfaces feature lets you describe what you need in a chat thread and have AI build the interactive report, dashboard, or presentation right there in the conversation. Which sounds like a feature that will mostly get used to build charts nobody asked for, but the underlying idea is interesting: shrinking the distance between the place where a question gets asked and the place where the answer gets visualized. For teams that already live in Slack, removing the "let me pull that up in another tab" step is a genuine quality-of-life improvement, and probably a modest but real time-saver across a workweek. The "vibe-code" framing in the headline is doing some heavy lifting; it's mostly just natural language prompting. But the output appears to be genuinely interactive, not just a static screenshot.
How AI Used Our Web Browser to Analyze Hundreds of LinkedIn Comments in Minutes
The Marketing AI Institute team used browser automation to scrape and analyze hundreds of LinkedIn comments at scale, surfacing themes and sentiment in a fraction of the time manual analysis would take. For senior marketers, the relevant application isn't necessarily LinkedIn comments specifically; it's the underlying method. The same approach works for review sites, community forums, competitor comment sections, anywhere audience language lives in the open. The setup requires some patience, but it's not a technical lift that requires a developer.
New AI Search & SEO KPIs: 4 Signals That Guide Real Decisions
The SEO measurement conversation has been having a bit of an identity crisis since AI Overviews started eating referral traffic, and this piece from LightSite AI's Stas Levitan is one of the more grounded attempts to sort it out. The argument is that the old KPIs, rankings, click-through rate, organic sessions, don't tell you much about how AI systems are evaluating and citing your content. The four signals Levitan outlines are first-party and connectable to human traffic, which makes them useful rather than just theoretically interesting. Worth reading alongside the Wikipedia story above, since the two pieces are describing the same problem from different angles.
Quick Hits
- Jensen Huang told investors Nvidia is projecting 70% growth next year; the circular-dealing concerns haven't gone away, but neither has the revenue. Read more
- UK GDP grew 0.4% in July and the services sector's AI expansion got partial credit, one of the first times a government stat has directly attributed economic growth to AI adoption, which either means something or gets heavily revised in Q3. Read more
- OpenAI launched a vertical-specific ChatGPT product for financial services, with built-in financial data and reasoning for research and modeling; the verticalization of foundation models is moving faster than most people expected. Read more
- Cognition, the agentic coding startup, is now valued at $48B, which suggests investors believe automated code generation is going to be large, or have made a decision they will revisit in 18 months. Read more
- GPT-6 Astra technical breakdown: looped transformers, recurrent depth mechanisms, hidden chain-of-thought reasoning, dense reading, but useful if you want to understand what's powering the tools you're using. Read more
- Wil Reynolds argues credibility now matters more than content volume in an AI-saturated environment, a reasonable point that is also, conveniently, reassuring for anyone who has been putting out less content lately. Read more
This article was obviously generated with AI but curated by a human, don't be weird about it.
Compiled by
Pete Bishop
Chief Innovation Officer, ZGM Modern Marketing Partners
Pete Bishop has spent the last two decades helping brands adopt new technology without losing the plot. He hosts Artificial Breakdown, a podcast and weekly newsletter that translates AI news into practical marketing decisions, and scans 50+ sources each week to write this issue.
Get this in your inbox
One email a week with the AI and marketing news worth your time.