Number one is a thing of the past
13 pageviews, Mon Oct 19 to Thu Oct 22
https://christianjward.com/blog/number-one-is-a-thing-of-the-past
Anthropic announced Claude Duet on Oct 21 as an assistant that comes to you first. The interface moves from you prompting it to the assistant watching your apps and bringing you what needs attention.
The announcement had about 49.8K likes and 7.5M views.
I've been running a small team of personal agents on Grok Bot, and each one has its own job. What usually makes one of them useful is how much it remembers about me and how accurate its information is about the people and businesses it deals with on my behalf.
The same holds for an assistant that brings me what needs my attention. The big test for me is how it handles personal channels versus work channels.
Many of us have multiple accounts, and it would be great to keep them all separate, directly inside one bot.
I do my rambling in Monologue voice notes, and now with Grok Bot too. Every night my Mac mini processes those notes, making connections, sorting topics and adding knowledge graph entries.
Morning is the fun part. Next I want to try Duet from Anthropic in that same overnight handoff.
https://x.com/claudeai/status/2112984664256241837
https://x.com/i/trending/2112987690803343606
FrontierMemory, from Letta, asks whether an agent's memory makes its next task better. It judges memory on learning from experience.
This week I posted that I'd love to see a ranking that scores how well an agent uses the context you give it. FrontierMemory judges the memory side of that.
Assistants now carry what they learned about you from one session to the next. For a business, what gets cited today shapes what gets recalled next week, and that compounds.
Agents working together need shared context and shared memory too. Without it, each agent starts from zero and repeats work another agent already finished.
https://x.com/Letta_AI/status/2112957074644873289
https://x.com/charlespacker/status/2112972810687823985
Wabi 2.0 is an agent the whole group chat shares. In her posts, @ekuyda argues the agent interface is "a messenger with your people in it."
That takes conversational AI from one-to-one chats into group chats.
I wrote about how conversation is rewriting what we expect from an interface.
https://www.christianjward.com/blog/conversational-interfaces-break-jakob-s-law
When the software can ask follow-up questions, the setup becomes a conversation. You describe the job and it works out the steps.
My own version is one voice talking to 20 other bots, and some of those bots are controlling Claude and some are controlling ChatGPT.
I can literally walk with it in my ear and just talk to it, go down to the kitchen, get a cup of coffee, and we're talking through 19 different projects at the same time. I don't think that's the average consumer.
https://x.com/ekuyda/status/2112212804979880239
https://x.com/ekuyda/status/2112278265884971446
In a post on marketing to AI agents, @SeijinJung wrote about Agent Conversion Optimization.
The new citation data comes from @Dmytro_Sokhach, who re-ran 11,025 "best X" queries from July to October. Google's citations of self-promo listicles, where a brand ranks itself, fell from 44% to 24%.
An AI model has to decide which facts to trust and which ones to repeat, and it leans on sources that agree with each other.
At Yext Research we looked at cases where every major AI model picked the same top business for a question. In 91% of those unanimous picks, the winning business's own website was one of the sources cited.
A brand with clean, consistent facts on its own site and everywhere else it shows up gets repeated. One with conflicting hours or an old menu gets passed over.
https://x.com/SeijinJung/status/2112259867851567188
https://x.com/Dmytro_Sokhach/status/2112241629281087513
Glenn Gabe posted that Walmart is opening to shopping agents while Amazon blocks them. In a separate post, @gregisenberg wrote about a "war on agents."
A blocked agent goes to the next source it can read and answers from there. For local businesses, those sources are mostly ones the brand already shapes.
Yext Research looked at 155.5M citations across 1,623 brands in location-grounded questions. 80% pointed to brand websites or business directory profiles.
That makes every blocked site a bigger test of the rest of your data. Your hours, menu, pickup options and prices have to match on your site, your directory profiles and every source an agent can still reach.
https://x.com/glenngabe/status/2113239935700308389
https://x.com/i/trending/2113059266378793184
https://x.com/gregisenberg/status/2111903028202852760
https://www.yext.com/research/facts/80pct-of-location-ai-citations-are-influenceable
AI now builds data-heavy, animated decks from real data, and it's insane how fast you can do this now. The animations have gotten so good.
I've been using Guess Who to explain game theory and search for a long time. What seemed to be science fiction is now everyday reality, and AI search is what we would call an infinite game, a term from James Carse.
It's a very different game. There is no number one anymore.
It changes all the time, and it's very much by market and by industry.
A broad question and a narrow local one need different strategies. The context at each level of the funnel affects all of our effort as marketers to target that level.
Google's got so much context that they don't tell you they have. Every query is gonna be so custom to the context and memory of the individual consumer.
It used to be you might still show up in search, and now in AI you're not going to because you haven't added all that additional content. You will if no one else has.
We are only going to win in AI if we give it absolutely every detail in real time that we can.