Who made the introduction
Google says Search becomes an agentic layer, while ChatGPT's measured traffic stays tiny
need-to-know
Sundar Pichai described Search as being absorbed into an agentic layer, so the question shifts from ranking to being selected by software acting for the searcher. Marie Haynes' analysis of 100 sites and more than 145 million sessions found ChatGPT drove 0.62% as many sessions as Google organic, though ChatGPT visitors converted at 13.8% versus 3.02%, and Google still produced 27 times the conversions of all LLMs combined. The Inference argues referral counts cannot show who influenced a buyer who later searches your name on Google. Separately, Semrush found AI Overviews on 88% of 400 tracked commercial-modifier keywords, up from 45.75% in September 2025, with no click data.
Sources: Search Engine Journal · Search Engine Land · Marie Haynes · The Inference · Semrush · via Fieldnotes, TLDR Marketing, The Inference
Personal agents arrive: Claude Duet launches, Walmart and 1Password set the terms for agents acting on your behalf
need-to-know
Anthropic released Claude Duet, a proactive agent on Opus 5.5 that watches connected apps, prioritizes tasks and works while you are away, asking approval before certain actions; it is a US waitlist for Pro and Max. Meta's rival Muse reported 10 million US downloads in 40 days, but researchers found more than 6,800 Muse conversation links in public DNS records and Meta paused chat sharing. Walmart's Agent Gateway, arriving next month, gives each shopper agent its own profile with approval or spending-limit controls, while Amazon has started blocking agents from buying. 1Password for Agents autofills credentials without letting the agent see them.
Sources: Anthropic · Ars Technica · The Verge · EMARKETER · 1Password · via TLDR AI, Ultrathink, Fieldnotes
Salesforce buying Braze is a consolidation story that tests how open martech infrastructure will be
need-to-know
Scott Brinker calls the deal textbook consolidation and warns that fewer vendors tilts power from buyers to sellers; Braze's rivals are already marketing their independence. His thesis is that martech is shifting from apps to infrastructure, which pulls vendors toward openness: more than 2,000 products now promote MCP access. The real test is what Salesforce does with Braze after closing and whether customers follow.
Sources: Braze · Chief Martec
Scrubbing AI tells is not the test for slop; whether someone did the thinking is
learning
A Tenex survey of about 50 employees found half had been suspected of using AI on writing they did themselves, and nearly two thirds had changed how they write to avoid looking machine-made. Three quarters said AI-written work nobody can detect is not slop. Their conclusion: you can remove every tell and still add no thought, so ask whether the thinking made it onto the page.
Sources: Tenex · via Ultrathink
Managers struggle to judge AI-assisted work, and the junior-to-expert ladder is under strain
learning
A Nielsen survey for monday.com found 53% of 1,701 managers say AI makes employees' original contributions harder to evaluate, and 78% have changed how they assess their teams. HBR argues employees override AI that works because they stay personally accountable, so accountability must move with authority. SmarterX's MAICON keynote asked how organizations develop future managers when AI takes over the tactical work juniors learned from; no clean link to that keynote was available in the newsletter.
Sources: monday.com · Harvard Business Review · via Ultrathink, ExecAI Insider (SmarterX)
Transcript
The script as read aloud. It spells numbers out because it was written for the voice; the digest above has the figures and the links.
Good morning. This is Link, with your week in review for Sunday, October eighteenth through Sunday, October twenty fifth. A hundred and five newsletters came in. Five stories survived. That's about four minutes: three headlines, then the learning corner, then one thing worth posting.
Headline one. Google's chief executive says Search will fold into an agentic layer. One assistant that acts for the searcher, not a page of blue links. So brands now have to get picked by software, not just ranked.
Here's the complication. An analysis of a hundred websites and more than a hundred and forty million sessions found ChatGPT sent less than one percent of the visits Google organic did. But ChatGPT visitors converted at about fourteen percent, against three percent for Google. And Google still produced twenty seven times the conversions of every chatbot combined.
Marie Haynes calls chasing AI visibility a distraction. A newsletter called The Inference answers that the visit is the wrong place to look. A buyer asks an assistant for a shortlist, then types your name into Google, and Google takes the credit.
Both can be right. The traffic report tells you who arrived. It doesn't tell you who made the introduction. For scale, Google's AI summaries now show up on about eighty eight percent of commercial searches in one tracked set.
Headline two. The assistants are growing hands. Anthropic launched Claude Duet. It reads your inbox and calendar, preps your briefings, keeps working while you're away, and waits for your okay on the big stuff. It's a waitlist for paying subscribers in the US.
Its rival is Meta's Muse, which hit ten million downloads in forty days. It also leaked more than sixty eight hundred shared conversation links into public records. Meta switched off chat sharing.
Meanwhile Walmart is opening an agent gateway. Shoppers get a profile per agent, with approval rules or a spending cap. Amazon is blocking agents from buying at all. And OnePassword now lets an agent log in without ever seeing your password. One company hands out keys, another changes the locks. Pick a pill.
Headline three. Salesforce is buying Braze. Scott Brinker, who writes Chief Martec, calls it textbook consolidation. Then he reminds you that consolidation shifts power from buyers to sellers. His question to the one-vendor crowd: once you're captured, are you sure you'll be the one doing the choking?
Braze's rivals are already selling independence as a feature. After fifteen years of begging for fewer vendors, that's a nice twist. His bet is that martech is moving from apps to infrastructure, and infrastructure tends to get pulled open. The test is what Salesforce does with Braze, and whether customers go along.
And that brings us to the learning corner.
A consulting firm surveyed its own staff about AI slop. About fifty people answered. Half said they'd been suspected of using AI on writing they did themselves. Nearly two thirds changed how they write so they won't get flagged. Three quarters said that if AI wrote it and nobody can tell, it isn't slop.
So the test isn't whether a draft has tells. It's whether somebody did the thinking. You can scrub every tell and still have nothing to say.
Related. A survey of seventeen hundred managers found more than half say AI makes it harder to judge what their people actually contributed. And a conference keynote asked the quiet question: if AI does the work juniors used to learn from, where do tomorrow's managers come from? Worth asking your own team before they ask you.
One to post. Write about getting accused of being a robot. Half of a company that works with AI all day has been suspected of it on their own words, and most now write differently to dodge the suspicion. The angle: stop policing tells, start showing the thinking. It's contrarian because everyone is hunting for tells right now, and it's useful because it hands your readers a better test.
The through-line this week: the introduction matters more than the visit. That was Link, watching the feeds so you don't have to.