Dan DeVries (@danieljdevries) header
Dan DeVries (@danieljdevries) profile picture

Growth, Ops, Sales & Marketing Executive | 30+ Yrs in Tech & Ecommerce, Last 7 in Home Decor & Accent Furniture | Tracking AI & Physical AI

Los Angeles
Joined February 2009
Photo by Dan DeVries (@danieljdevries)Photo by Dan DeVries (@danieljdevries)Photo by Dan DeVries (@danieljdevries)Photo by Dan DeVries (@danieljdevries)Photo by Dan DeVries (@danieljdevries)Photo by Dan DeVries (@danieljdevries)Photo by Dan DeVries (@danieljdevries)Photo by Dan DeVries (@danieljdevries)Photo by Dan DeVries (@danieljdevries)Photo by Dan DeVries (@danieljdevries)
Tesla's FSD v15 release notes say the new vision encoder is shared across FSD, Robotaxi and Optimus, trained on millions of internet videos, and improves the critical intervention rate by 25x. That shared stack is the real Optimus story: every mile the fleet drives sharpens the eyes of a humanoid robot headed for warehouse aisles and kitchens. On a DC floor, intervention rate drives pick and pack costs more than raw speed does. The most real-world data wins. xitter.cf/TeslaSI/status/2124229…
Full release notes below – – – v15 upgrades every part of the end-to-end driving network. Includes: Upgraded vision encoder shared between FSD, Robotaxi, and Optimus, which is trained on millions of internet videos for improved generalization and data efficiency, improving the critical intervention rate by 25x. Introduced Tesla Reality View, replacing all of the explicit model signals into a single end-to-end visualization using less total compute. Improved navigation by using an end-to-end learned model based on route preference data from the fleet. Improved parking location selection by using personalized parking history, and the map will now display multiple parking options at your destination. Improved driver monitoring by increasing vision-based attention monitoring availability with hats and sunglasses and decreasing strictness in higher speed driver profiles. Improved detection of road signs, small animals, road debris, and chains. Improved detection of potholes, road details, and avoidance of deep standing water. Improved speed handling for each driver profile, school zones and speed limits, and inclement weather. Improved Autoshift functionality by unifying with FSD-based gear predictions.
2
67
Replying to @AlanBin44785396
They should. "Looked smaller in person" or "finish didn't match" tells a recommender more than a star rating. It's a big reason we make the main image a lifestyle shot, so buyers can judge scale and finish in a room before anything ships.
1
4
Replying to @ChaoXu457633
Good catch on the budget check. Testing the cap before the payment token is issued is a different control from a log you read afterward. One stops the overspend, the other just documents it.
3
Replying to @kcarrillo10
Optimus is the part I'll be watching. Humanoid robots in warehouses and homes are the bigger long-term story, and anything on factory deployments or pick and pack costs will say more than another demo.
5
Replying to @michaelchen
That matches what I see from the supplier side. Most retailers are still on FAQ bots and AEO tests, while the harder work is getting clean, complete product data from vendors so an agent can trust it enough to recommend and buy.
5
Replying to @LogicalSG
Trailer loading is still one of the most manual jobs in a furniture DC, and bulky, odd-shaped cartons make it harder. A pallet truck that works the dock without fixed safety infrastructure is the kind of retrofit an existing building can actually absorb.
4
Replying to @twitFromRup
Approve-each-change is the correct default here. Pricing and inventory edits are where an agent can do real damage fast, so read access first and gated writes is how sellers will build trust that it's acting in their interest.
10
Replying to @radio213
Product data is where this gets won. An agent can only recommend what it can read, so titles, bullets, attributes, romance copy, and reviews and other UGC have to hold up on their own. In home decor that means real dimensions, finishes and materials, not just a nice lifestyle shot.
1
1
9
Waddle Labs says its new Waddle-1 agent scored 81 in Robocurve's evaluation of two-armed robots doing unseen tasks zero-shot, ahead of GPT-6.2 Astra at 65. The design matters as much as the score: a reasoning model in the cloud directs fast policies running locally on the robot. In a warehouse, that split decides whether a robot can handle a new item without retraining and still move fast enough to lower pick and pack costs. Handling new items is the bar. Doing it at human speed is the business case. xitter.cf/theWaddleLabs/status/2…
We're announcing our first robot-use agent, Waddle-1. It controls bimanual robots on unseen tasks zero-shot. Our agent is the most capable system evaluated by @robocurve, scoring 81 compared to 65 for GPT-6.2 Astra and 59 for Dream-1. It calls LLMs in the cloud to direct lower-level policies that run locally. Waddle-1 is a first step towards agents that combine the capability of reasoning models and the efficiency of reactive policies. We expect such systems to achieve human-speed manipulation in the near future. If you would like to deploy robots with our preview API, please tell us what you want to build here: waddlelabs.ai/api-access
32
Salesforce's State of Agentic AI in the Enterprise report finds 39% of companies running agents already use AI to build more agents, but only 32% have a central system for managing them. Companies with stronger governance caught agents operating out of bounds 68% of the time, versus 55% for the rest. Every agent needs an owner, a defined scope of access and an audit trail before it goes live, the same as a new hire. People will only trust agents that are accountable for acting in their interest. xitter.cf/salesforce/status/2123…
Companies create AI agents faster than they can keep track of them 39% of organizations running agents already use AI to build more agents. Another 44% are testing it. But only 32% have a centralized system for managing agents. That’s where things can get messy. You need to know what each agent can access, what it can do, and who owns the outcome when something goes wrong. Governance often catches up after launch. Organizations with stronger governance caught agents operating out of bounds proactively 68% of the time, versus 55% for the rest. Read the State of Agentic AI in the Enterprise: sforce.co/4AF9F3H
5
2
72
NASASpaceflight reports SpaceX may now fly Starship Flight 16 from Starbase with Booster 23 and Ship 43, which would push the first LC-39A launch to February or later. The detail I'd watch: Ship 44 reached cryo testing seven days sooner than Ship 43, and SpaceX is building at least one booster a month. Henry Ford's moving assembly line taught the same lesson: the first car proved the idea, and the line made it a business. Reuse only pays when production keeps up, and that's how the cost per pound to orbit comes down. xitter.cf/NASASpaceflight/status…
2
1
3
111
The Walmart connector instructions for Meta's Muse split the work between two agents: Muse searches and changes the cart through Walmart's MCP, while Sparky, Walmart's own agent, answers questions on orders, stores, policies and comparisons and changes nothing. One detail matters from the supplier side: Walmart's cart keeps a product photo only when the shopping agent passes the image along. When agents hand off to agents, the images, titles and attributes have to survive every step to the shopper. xitter.cf/juokaz/status/21241717…
Walmart and Muse are about to launch the first agent-to-agent agentic commerce integration. Muse will use Walmart's MCP to get products, create carts and place orders, but it will talk to Walmart's Sparky to get information on "orders, stores, policies and comparisons."
33
The Philippines is in early talks with SpaceX to host commercial launch sites, and President Marcos says the goal is the launch-site business, not a national space program. Geography is a cost line: Europe launches from Kourou in French Guiana because a site near the equator with open ocean to the east gets more help from Earth's rotation and a safer flight path. As launch cadence climbs, pads and range time become the constraint. More sites with room to fly often is how the cost per pound to orbit keeps falling. xitter.cf/pnagovph/status/212368…
SpaceX in PH. The Philippines is exploring a potential business partnership with billionaire Elon Musks’s SpaceX to develop rocket launch sites in the country, President Ferdinand R. Marcos Jr. says on Friday (Nov. 20, 2026), although he clarifies that discussions remain at an early stage. The President said the initiative is focused on positioning the Philippines as a commercial launch site for SpaceX, not on establishing a national space program. He also cited the Philippines’ location along the western edge of the Pacific could make it a suitable site for rocket launches. | Darryl John Esguerra
2
288
Replying to @bernardocore3
On the seller side, that's where it lands. A marketplace fee is easier to plan around than interchange, but suppliers will want to see it per order, the same way they already track retailer chargebacks and drop-ship fees. Repeat B2B orders for the same items look like the first place agents take over.
15
AWS Marketplace now lets AI agents buy products on their own, paying per request in US dollars on the customer's existing AWS bill, with spending limits, approved-seller lists and every charge logged in CloudTrail. That is the plumbing agentic commerce needs before it reaches retail at scale: the buyer sets the rules, and the agent can show it followed them. Agents earn trust when they spend inside the owner's limits and act in the owner's interest, not for whoever paid for placement. https:/ x.com/awswhatsnew/status/212…
AWS Marketplace now supports agentic payments with AgentCore AWS Marketplace now supports agentic payments over the x402 protocol, so AI agents can discover and buy select products and pay per request in US dollars through AgentCore payments, part of Amazon Bedrock AgentCore....
4
3
162
Replying to @AlanBin44785396
Return rate would settle it. A bigger basket that comes back in a box isn't a win for anyone, least of all the supplier paying freight both ways.
1
3
Replying to @francescocrive
The test I'd use is whether the robot runs the same task on a live shift, every shift, with the customer paying for the output. In a warehouse that means owning pick and pack costs, not shipping another demo of a new skill.
4
Replying to @akash_dongre
Same logic we already use with warehouse automation: the machine can be smart, but the hard stops live outside it. For humanoid robots working around people and inventory, an independent layer that can refuse an unsafe action is what will let operators trust them on the floor.
1
27
Replying to @MrSharonSolomon
For retailers the answer has to be the product page. If the titles, bullets, romance copy, attributes, photos and videos are thin, an agent will summarize the product badly or pass on it, and reviews and other UGC end up telling the rest of the story.
25
Replying to @jahangirdanish_
Commerce has the same gap. If an agent shops through my logged-in session, nobody can tell whether I picked the product or the agent did, and people will only hand that over when they trust the agent to act in their interest.
2
2
10