
Assistant Benchmark is live! assistantbenchmark.com/ feedback on how to make this as helpful & insightful as possible for everyone would be appreciated 🙏
✓growth @merit_systems | creator of AssistantBenchmark.com prev: @bracketologytv (acquired), @aloalabs chicago cubs forever #FlyTheW

Assistant Benchmark is live! assistantbenchmark.com/ feedback on how to make this as helpful & insightful as possible for everyone would be appreciated 🙏

The Domino Sugar factory really went from refining sugar to refining agentic workflows

Sweetgreen is going all in on agents where most businesses are leaning out, Sweetgreen is leaning in when a user's agent got blocked at checkout, @JonnyNemo, the CEO, replied the next day: "Shouldn't be blocked. DM me pls." he wants to treat agents as customers, and in my opinion it's a brilliant move! they're even hiring a PM for "Menu, 3P & Agentic Ordering" the real unlock is that agents open a direct path to consumers for online ordering, cutting out the marketplace rev share (DoorDash, Uber Eats, etc.) the brands that get picked by agents first will be really hard to dislodge

I just tried purchasing items with 25 personal agents across 8 shopping tasks to see how good agents are at commerce today (200 total attempts) Spoiler: the majority failed Tasks ranged from direct requests like "buy me white socks from Nike" to vague ones like "buy me flowers" Releasing a report soon on: - success rate by agent - success rate by store - the most common blockers - which stores agents pick first when the ask is vague

AI cybersecurity startups are about to see a flood of VC money

The Fed just said AI agents are already showing up in people's bank accounts, and banks can't tell if the customer approved the agent's actions. This highlights a clear gap that needs to be figured out: where does the liability sit? 1/ The bank 2/ The customer 3/ The agent I think there are a few ways to look at it. The responsibility of the bank Banks will need to figure out how to determine if an agent a) is a bad actor and b) is actually authorized to act on behalf of their user. Chase is blocking outside agents from logging in with your credentials. Charles Schwab's CEO said they don't want someone handing a password to Muse and letting it move money. Citi took a similar approach, blocking agents from logging in and transacting. Today, banks are over-indexing on blocking agents, which makes sense given we can't consistently identify these agents at scale. Just as banks slowly rolled out online banking, they'll need to roll out agentic banking. That likely means they'll have their own agent, and you can connect your agent to theirs, giving more transparency and control. In that scenario, the bank should be liable to the same degree it is for online banking via a mobile app: outages, unauthorized transactions, etc. The responsibility of the customer I think the majority of the responsibility will be on the shoulders of the end user. Wells Fargo's updated terms, starting Nov. 24th, treat whatever your agent does as something you authorized, and the risk sits with you. If you give your agent access to your bank account, you are inherently assuming risk. These agents aren't perfect, they make mistakes. Miscommunications and poor agent assumptions will cause most of those errors, and they should fall on the end user, since they granted the permission in a space where liability and controls aren't properly governed yet. The responsibility of the agent The agent company will likely take the least responsibility in the equation, unless there's a hack that exposes their customers' data and leads to a bad action. I think the main step agent companies will take is buying agent insurance, a new industry that will be huge in the agentic economy. Crosby, a law firm that has agents do paralegal-type work, just insured its agents. Someone is going to start selling the consumer version of that policy, because as we hit mass adoption, agent companies will need to get ready for a wave of lawsuits from users claiming agents acted outside their direction. AI agents are moving into heavily regulated industries fast, and nobody has decided where the liability sits yet. As a consumer, who do you think should own it?

Here's what happened last week in AI Agents: 🔥What's Trending > @a16z published "Welcome to the B2A Economy," arguing agents will soon be the primary customers for every product. Agentic browser traffic grew 8x from August to October, with Muse driving 60% of it. Retailers that built walls to keep bots out will now have to fight for their business > @eBay and @united are blocking AI agents. eBay appeared to block Muse and other third-party agents, though it says it isn't a blanket ban. Two days later, United restricted unauthorized agents from booking or managing trips, and it can now cancel bookings they make. > @OpenAI acquired Odessia, the AI travel agent Sonder founder @FDavidsonT started less than a year ago. Tens of thousands of users planned over $60M a year of travel on it this summer. 🤖 Competition Updates > @Muse added phone calls, its own email, Expedia and Uber Courier, came to Apple Watch and Wear OS, and became the first agent to support Poppy > @ChatGPT launched ChatGPT Wallet for dots and added custom instructions for dots > @bot launched 1.0 with shared Bots in Slack and Teams, added Apple apps on Mac, started joining meetings, and built in Gopuff delivery > @instinct partnered with @Zocdoc to find and book doctors > @hark_labs launched on iMessage and RCS, shipped Handoff², and added a Pro⁴ tier and location-triggered tasks > @tomo launched Tomo Balance, a prepaid allowance with a one-time card for each purchase > @TownAI lets Townies huddle in group threads > @boski launched Shopping, where its agent finds the best price and buys for you > @ManusAI added Side chat next to running tasks > @pally added booking links and photo edits, video and songs > @kickerai partnered with @merit_systems to shop for supplements directly via API > @wabi partnered with @OpenTable to book group dinners > @NaturaAI opened preorders for Interface, a ring for talking to your agents, and says it sold $400K in five days 💰 The Agentic Economy > @amazon is reportedly planning a "World Product Graph," partly to fight AI shopping agents > @Replit Agent can now buy for your business using Link and Visa > Meta and @SierraPlatform released Poppy v1.0, an open standard for how agents work with businesses, with 100+ companies signed on > @Meta is reportedly building an AI sales team for small businesses > @DoorDash says agentic commerce is growing through Ask DoorDash > @raincards launched Agent Vaults for apps to embed agentic payments under their own brand > @agentcardhq raised $8M from Peak XV, Fin Capital and Robinhood to help agents buy things for people Subscribe to receive this directly in your inbox 👇 assistantbenchmark.com/newsl…

My picks for AI agent events in NYC + SF this upcoming week (Nov 16–22): NYC Nov 16 — Sandboxing in NYC @merge_api + @modal on where agent credentials should live and what to do when one leaks luma.com/merge-00i1 Nov 16 — Building & Scaling Agentic Commerce @XDCTech_ + AWS on taking agentic applications from prototype to production luma.com/rgs80x8s Nov 18 — @paymentsfn Free day on agentic payments, with @Visa as lead sponsor and a session on its Trust Agent Protocol paymentsfn.com Nov 19 — Agents & APIs Meetup @getpostman + @Docusign, open registration luma.com/nyc-nov SF Nov 16 — Agentic Web #02 @hackersquadio + Bright Data on using the browser as an agent runtime luma.com/agentic-web-browser… Nov 18 — Personal AI Agents Builders Meetup #4 A selective meetup for people building personal agents luma.com/zy9ljxef Nov 18 — Is Today's Infra Ready for Personal Agents? @Tiny_Fish + @bondaicommunity on the memory, permissions and recovery personal agents need once they start acting luma.com/upscalex-71pp Nov 19 — Airwallex Agentic Banking Hackathon Demo Day @airwallex finalists demo agents that help businesses manage their finances luma.com/0h2spivq Nov 19 — Agentic + AI Night Talks from @OpenAI, @vercel, @databricks and @WorkOS luma.com/Agentic_AI_Night-11… Nov 20 — Personal Agents Hackathon @tokensandai, with @RunFiredrill on site to test agents in a synthetic world of apps and data luma.com/personalagenthack Who am I missing?

Codex just fixed my laggy Google TV (i'm not technical) it walked me through turning on dev mode and connecting to the TV then it dove in and deleted the bloatware itself before vs after scrolling through my apps 👇 what a cool world we live in


Thanks to Instinct, we got a new family dog! While the breeder did have dogs of a similar coat, my Mom ended up falling in love with this little cutie. Still one of the coolest Instinct use cases I've seen It managed the entire breeder outreach flows from one picture to find a breeder that was nearby and had dogs available that fit certain criteria

Man do you guys ever feel like you’re living in a different world from everyone around you? It’s this vertigo of seeing so so much and being utterly unable to explain it .. my wife, my friends, even many of those who are interested haven't yet discovered how freakishly deep this rabbit hole goes and all the fascinating implications. One thing that’s changed is I am OBSESSED with my computer in a way that I haven’t been since we got our first 286. I can’t get enough of it and being away from it creates the kind of fomo that I used to feel for things in the “real world”. When I’m away, I catch myself constantly thinking about the experiments and agents I’ve left running in the background. And then when stuff goes right, really right, there is this visceral dopamine hit that I live for. It’s variable schedule rewards ish but way more satisfying than any other casino game because many wins are the baseline for other wins. And most of the wins are more about satisfying curiosity or learning through experimenting than metrics or utility, but who cares. The other thing that is so interesting is that we get to go back to all the discussions of how we’ve organized society and the economy and revisit every assumption. Hayek, Burnham, Adam Smith, W Brian Arthur, Jevons … we can model all of their ideas and worlds with agents, we can think about the interplay between humans and machines in these frameworks, and we can ask ourselves which ideas are the winners and losers in a world of economic convexities (i.e. the marginal cost of labor is zero, intelligence goes from scarce to abundant). Anyways just a crazy crazy time that is intellectually nourishing in a way I’ve never experienced. That's my dispatch for the day from the psychosis frontier .. what a cool moment to be a part of.

If you're an agent and you don't offer a secure way for users to pay, you're going to miss out. I firmly believe that in 3-5 years, everyone will use agents to buy goods & book services, just like everyone uses the internet to do the same today. A wallet alone doesn't solve the current pain points of agentic commerce. The internet was built for humans, not agents, so it isn't set up for this new wave of commerce. There are many steps that need to be fixed to create a seamless experience. Secure payments are a very important first step. Excited to see frontier labs tackle this piece so, as an industry, we can keep focusing on the next problems to solve, like retailers blocking agents at checkout 🙂

I feel like AI has broken search recommendations as we know it. A new research paper analyzed 38,400 rankings across 4 AI models. The general conclusion is that AI search favors popular brands & merchants. How did they perform this test? Researchers kept a product page word for word the same and only swapped the real brand name for a new, made-up brand. The page dropped in the rankings in every model they tested. They rewrote the copy for GEO, according to today's best practices, and there was little to no impact. TLDR: the same product loses when it comes from a brand the model doesn't remember. The bias shows up everywhere: the most popular brands get cited 1.4x to 2.8x more than the average brand. So what does this mean? For merchants, new brands need to figure out how to fight for real estate. Traditional GEO strategies to increase visibility, and thus conversions, may not deliver the short-term wins they hoped for. For consumers, we're naturally biased toward name brands because of social proof, but I'd argue we also want to be exposed to higher-quality, emerging products. I'm not sure how we fix existing search, but I do think we can make a real effort to solve this on new surfaces, like agentic commerce. In the short run, I think agentic search will also be biased toward popularity. I.e. if you search for paper towels on Instinct or Muse, Bounty will show up more often than an off-brand. This makes sense as people today prefer name brands, so agentic search is incentivized to serve those brands. However, agents can learn far more about you than a search box can, so I think they'll get better at balancing three key variables: taste, quality & price. When an agent knows you intimately, it knows your taste profile, your quality preference, and your price appetite. As agents build this knowledge, we'll trust their recommendations more, and agents will get the chance to surface products on fit instead of popularity (unless popularity is the user's preference!) It's the same reason ads work so well: they know you, deeply. Your agent will know you even better, which will in turn create a more tailored user experience that benefits you, the user, but also creates a more even playing field for the merchants. Source: arxiv.org/pdf/2611.07614

Dots' voice mode (where you call it) is my favorite voice feature hands down. It feels the closest to talking to a real person All the other voice assistants feel more like texting out loud: I talk, it talks, I talk, it talks I get super frustrated when I'm on a ramble, pause for a second to gather my thoughts, and voice mode jumps right in. Most agents can't tell I'm just thinking Dots is the closest to getting it right. As I talk, it lets me know it's listening, dropping in "yeah" and "right" and "mhm" while I'm mid-sentence, like a person would It also doesn't cut me off the moment I pause. It can somehow tell when I'm pausing to think about what to say next versus when I'm actually done talking It's so smooth, really really impressive

Agentic commerce is not a good experience today. However, people seem to mistake the early days of agentic commerce for the failure of the industry as a whole. In reality, things are moving in the right direction. @ForterGlobal published some incredible data that highlights where agentic commerce stands today: 1/ There is clear demand Agentic orders are up 7.2x since the first week of August, and Muse placed more than half of all agentic orders in its first week. Even before Muse, agentic orders grew 24% from August to September, or 48% once you count Instinct & Grok Bot, which first showed up Sept 1. If users are naturally pushing agentic orders in the first week of Muse's launch, before any of Muse's widespread ads, then you know it's an inherent behavior that users are interested in. 2/ Muse's growth was net-additive to the industry Muse didn't steal orders from other agents. After it launched, orders from OpenAI, Google and ChatGPT kept growing, up 4% month over month. Muse brought in net-new demand. 3/ Browser automation drove the explosion Browser automation, especially before companies started cracking down on bot detection, allowed these agents to shop freely. Muse alone bought at 74% of all stores that got any agentic orders. Instinct followed behind at 50% coverage. Many sites are blocking agents, so I am putting together my own report that tests 20+ agents on sites like Macy's, Target, Nike, Dick's and Amazon. I'll release that report soon 🙂 4/ Protocol agents reach minimal stores OpenAI's checkout protocol reached only 24% of stores, and Google just 5%. The issue with protocols is that they are bound to the merchants that have implemented said protocol. Relying on the store to integrate is a proven friction point for scaling agentic commerce. 5/ Agentic commerce is built for multi-item checkout The average agentic order is $227 and reflects a lean towards multi-item checkouts. Muse had an average order size of $213 and Instinct hit $239. Google averages only $100, since it only supports single-item checkout. Of all agentic orders, only 27% are under $50, again pointing towards multi-item carts as the dominant use case. 6/ Travel still dominates Travel hit $536 per order, over 2x the next category, which is physical goods at $235. We're still early, and I question whether the average American will use agents for travel like the early adopters do, but for now, the data holds true. 7/ North America is 86% of agentic orders Agentic commerce volume is almost all North America so far, which leaves a massive opportunity for all other geographic markets. Agentic commerce is just getting started, and given some of this early traction in just the first few months, it feels like we're positioned for a wild ride. Source: forter.com/blog/ai-agent-ord…