Tech
Designing the agentic AI enterprise for measurable performance
Presented by Edgeverve
Smart, semi‑autonomous AI agents handling complex, real‑time business work is a compelling vision. But moving from impressive pilots to production‑grade impact requires more than clever prompts or proof‑of‑concept demos. It takes clear goals, data‑driven workflows, and an enterprise platform that balances autonomy, governance, observability, and flexibility with hard guardrails from day one.
From pilots to the “operational grey zones”
The next wave of value sits in the connective tissue between applications — those operational grey zones where handoffs, reconciliations, approvals, and data lookups still rely on humans. Assigning agents to these paths means collapsing system boundaries, applying intelligence to context, and re‑imagining processes that were never formally automated. Many pilots stall because they start as lab experiments rather than outcome‑anchored designs tied to production systems, controls, and KPIs.
Start with outcomes, not algorithms. Translate organizational KPIs (cash‑flow, DSO, SLA adherence, compliance hit rates, MTTR, NPS, claims leakage, etc.) into agent goals, then cascade them into single‑agent and multi‑agent objectives. Only after goals are explicit should you select workflows and decompose tasks.
Pick targets, then decompose the work
What does “target” actually mean? In agentic programs, a target is a business outcome and the use case that moves it. For example, “reduce unapplied cash by 20%” target outcome; “cash application and exceptions handling” use case. With the use case in hand, perform persona‑level task decomposition: map the human role (e.g., cash applications analyst, facilities coordinator), enumerate their tasks, and identify which are ripe for agentification (data retrieval, matching, policy checks, decision proposals, transaction initiation).
Delivering on those tasks requires a data‑embedded workflow fabric that can read, write, and reason across enterprise systems while honoring permissions. Data must be AI‑ready, discoverable, governed, labeled where needed, augmented for retrieval (RAG), and policy‑protected for PII, PCI, and regulatory constraints.
Integration goes beyond APIs
APIs are one mode of integration, not the only one. Robust agent execution typically blends:
-
Stable APIs
with lifecycle management for core systems
-
Event‑driven triggers
(streams, webhooks, CDC) to react in real time
-
UI/RPA fallbacks
where APIs don’t exist
-
Search/RAG connectors
for documents and knowledge bases
-
Policy management
across tools and actions to enforce entitlements and segregation of duties
The north star is integration reliability — built on idempotency, retries, circuit-breakers, and standardized tool schemas — so agents don’t “hallucinate” actions the enterprise can’t verify.
A quick example: finance and facilities, in production
Inside our organization, we deployed specialized agents in a live CFO environment and in building maintenance. In finance, seven agents interacted with production systems and real accountability structures. Year‑one outcomes included: >3% monthly cash‑flow improvement, 50% productivity gain in affected workflows, 90% faster onboarding, a shift from account‑level handling to function‑level orchestration, and a $32M cash‑flow lift. These results don’t guarantee gains everywhere; they show that designing products can deliver measurable outcomes on a scale.
The four design pillars: Autonomy, governance, observability & evals, flexibility
1) Autonomy: right‑size it to the risk
Autonomy exists on a spectrum. Early efforts often automate well‑bounded tasks; others pursue research/analysis agents; increasingly, teams target mission‑critical transactional agents (payments, vendor onboarding, pricing changes). The rule: match autonomy to risk, and encode the operating mode suggest‑only, propose‑and‑approve, or execute‑with‑rollback per task.
2) Governance: guardrails by design, not as bolt‑ons
Unbounded agents create unacceptable risk. Build guardrails into the plan:
-
Policy & permissions: tie tools/actions to identity, scopes, and SoD rules.
-
Human‑in‑the‑loop (HITL): where mission‑critical thresholds are crossed (amount, vendor risk, regulatory exposure).
-
Agent lifecycle management: versioning, change control, regression gates, approval workflows, and sunsetting.
-
Third‑party agent orchestration: vet external agents like vendors, capabilities, scopes, logs, SLAs.
-
Incident and rollback: kill‑switches, safe‑mode, and compensating transactions. This is how you
scale innovation safely while protecting brand, compliance, and customers.
3) Observability & evaluations: trust comes from telemetry
Production agents need the same rigor as any core platform:
-
Telemetry: capture full execution traces across perception, planning, tool use, action supported by structured logs and replay.
-
Offline evals: cenario tests, red‑teaming, bias and safety checks, cost/performance benchmarks; baseline vs. challenger comparisons.
-
Online evals: shadow mode, A/B, canary releases, guardrail breach alerts, human feedback loops.
-
Explainability & auditability: why was an action taken, which data/tools were used, and who approved.
4) Flexibility: assume volatility, design for swap‑ability
Models, tools, and vendors change fast. Treat agentic capability as platform currency: create an environment where teams can evaluate, select, and swap models/tools without tearing down the build. Use a model router, tool registry, and contract‑first interfaces so upgrades are controlled experiments, not rewrites.
The agent platform fabric: how platformization turns goals into outcomes
A true agentic enterprise requires a platform fabric that transforms goals into outcomes, not a patchwork of isolated pilots. This platform anchors enterprise‑to‑agent KPI cascades, drives task decomposition and multi‑agent planning, and provides governed tooling and data access across APIs, RPA, search, and databases.
It centralizes knowledge and memory through RAG and vector stores, enforces enterprise controls via a policy engine, and manages performance and safety through a unified model layer. It supports robust orchestration of first‑ and third‑party agents with common context, embeds deep observability and evaluation pipelines, and applies disciplined release engineering from sandbox to GA. Finally, it ensures long‑term resilience through lifecycle management versioning, deprecation, incident playbooks, and auditable histories.
Guardrails in action: a BFSI example
Consider payments exception handling in banking — high stakes, regulated, and customer‑visible. An agent proposes a resolution (e.g., auto‑reconcile or escalate) only when:
-
The transaction falls below risk thresholds; above them, it triggers HITL approval.
-
All policy checks (KYC/AML, velocity, sanctions) pass.
-
Observability hooks record rationale, tools invoked, and data used.
-
Rollback/compensation is defined if downstream failures occur. This pattern generalizes to vendor onboarding, pricing overrides, or claims adjudication — mission‑critical work with explicit safety rails.
Scale beyond pilots
Scaling agentic AI beyond pilots demands disciplined readiness across nine fronts: leaders must clarify which KPIs matter and how agent goals ladder into them, determine which persona tasks are agentified versus remain human‑led, and align each with the right autonomy mode from suggest‑only to propose‑and‑approve to execute‑with‑rollback. They must embed governance guardrails, including HITL points and lifecycle controls; ensure robust observability and evaluation via telemetry, replay, audits, and offline/online tests; and verify data readiness, with governed, policy‑protected, retrieval‑augmented data flows. Integration must be reliable, with API lifecycle management, event triggers, and RPA/other fallbacks. The underlying platform should enable model swap‑ability and orchestration of first‑ and third‑party agents without rebuilding. Finally, measurement must focus on true operational impact cash flow, cycle times, quality, and risk reduction rather than task counts.
The takeaway
Agentic AI is not a shortcut; it’s a new system of work. Enterprises that approach it with platform discipline aligning autonomy with risk, embedding governance and observability, and designing for swap‑ability will convert pilots into production impact. Those that don’t keep accumulating impressive but disconnected demos. The difference isn’t how fast you ship an agent; it’s how deliberately you design the enterprise around it.
N. Shashidar is SVP & Global Head, Product Management at EdgeVerve.
Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact sales@venturebeat.com.
Tech
Microsoft admits Game Pass may have gotten too expensive
![]()
An internal memo obtained by The Verge shows Xbox’s new CEO, Asha Sharma, acknowledging that Game Pass has grown too expensive. Microsoft is now exploring lower-cost subscription tiers, though no firm timeline has been set.
Read Entire Article
Source link
Tech
Blackmagic's camera for Apple Vision Pro content is now available for $30K
Blackmagic’s URSA Cine Immersive camera for Apple Vision Pro content can now be ordered, if you have $30,000. Here’s what it can do, and who it’s for.

Blackmagic’s URSA Cine immersive camera is here.
Announced back in June 2024, the high-end camera was presented as an end-to-end workflow for the Apple Immersive Video format used by the Apple Vision Pro. Apple Immersive Video delivers 8K 3D video with a 180-degree field of view, along with support for Spatial Audio.
With the Apple Vision Pro, Apple delivered immersive 3D video experiences, giving wearers the chance to re-live their favorite moments with the help of formats such as Spatial Video. Exclusive content, in a special format called Apple Immersive Video, is also offered on the headset.
Continue Reading on AppleInsider | Discuss on our Forums
Tech
OpenAI sees ‘staggering’ demand for Amazon offering, says Microsoft partnership held it back

OpenAI’s partnership with Microsoft helped launch the generative AI era. Now Amazon is bringing the ChatGPT maker further into the booming market for enterprise AI.
That’s the takeaway from an internal memo distributed over the weekend by Denise Dresser, OpenAI’s chief revenue officer. The memo touted the Amazon Web Services alliance as a key enterprise growth driver for the ChatGPT maker, according to a CNBC report.
“Our Microsoft partnership has been foundational to our success. But it has also limited our ability to meet enterprises where they are — for many that’s Bedrock,” Dresser wrote, referring to the AWS AI model platform. “Since we announced the partnership at the end of February, inbound demand from our customers for this offering has been frankly staggering.”
It’s a high-profile endorsement for AWS, the leading cloud platform by market share, and perhaps Amazon’s best answer yet to the persistent perception that it was caught flat-footed by the generative AI boom sparked by the launch of ChatGPT in November 2022.
The OpenAI-Amazon relationship has come full circle. As GeekWire previously reported, AWS was actually OpenAI’s first cloud partner, providing computing resources at the lab’s founding in 2015, before Microsoft swooped in.
A decade later, Amazon and OpenAI in February struck a $50 billion investment and a cloud deal worth more than $100 billion over eight years.
Microsoft’s deal with OpenAI remains intact. Azure is still the exclusive host for OpenAI’s core APIs, and Microsoft retains its intellectual property license and revenue-sharing arrangement, including a share of revenue from OpenAI’s partnerships with other cloud providers.
Dresser’s memo also took aim at Anthropic, whose Claude model has emerged as the popular favorite in enterprise applications and software development. She said Anthropic made a “strategic misstep to not acquire enough compute,” echoing a separate OpenAI memo to investors that characterized Anthropic as “operating on a meaningfully smaller curve.”
Amazon also has a major investment in Anthropic, having committed $8 billion to the Claude maker. Both Seattle-area tech giants now hold stakes in the two leading AI labs, showing how quickly the era of exclusive AI partnerships has given way to a messier set of alliances.
Tech
Huawei Teases a Wider Foldable, and the Timing Feels Very Apple-Adjacent
Huawei’s new extra-wide foldable phone was revealed on Monday, and it’s already drawing comparisons to the next big thing from Apple. Little is known about the Pura X Max, teased on China’s Weibo social network, but the internet is buzzing over its uncommon proportions, wide aspect ratio and similarities with the rumored iPhone Fold expected to arrive later this year.
There are plenty of different ways to fold a phone, and tech companies have been experimenting with various configurations over the last few years. There are clamshell-style devices, like the Motorola Razr and Samsung Galaxy Z Flip 7, which fold like OG flip phones, shrinking down into a pocketable square. Then came taller, book-style phones that unfold to reveal larger, tablet-size displays. These days, you can even find trifold phones with two hinges. However you fold it, most foldables either start or end with the familiar candybar shape as their base.
Manufacturers have experimented with a variety of different foldable configurations, yet the Pura X Max’s wide proportions are uncommon.
Huawei’s Pura X Max — the newly announced larger evolution of the Pura X foldable phone — is what I like to think of as a passport-style foldable. Rather than the tall rectangle aspect ratio that almost all modern phones take, it has shorter and extra-wide proportions when folded before opening up to reveal what appears to be an approximately iPad Mini-sized and proportioned internal display. This is a formula similar to what we’ve seen on the 2020 Microsoft Surface Duo, though the Duo made use of dual internal screens split by a visible hinge rather than a flexible, foldable display.
The proportions of the Pura X Max announcement have experts and enthusiasts drawing comparisons to the rumored iPhone Fold, which might end up being called the iPhone Ultra, that is expected to debut this fall alongside the new iPhone 18 models. Industry insiders, alleged dummy models and leaked CAD files point to Apple using a similarly wide passport-style shape with proportions in the ballpark of the Google Pixel 10 Pro Fold. The iPhone Ultra is rumored to feature a 5.5-inch outer screen that unfolds into a 7.8-inch internal display and could be priced between $2,000 and $2,500.
Apple’s rumored foldable iPhone is also expected to use a wide aspect ratio, though perhaps not this wide.
Set to launch on April 20 in China, the Pura X Max is shown in Huawei’s teaser imagery sporting a triple-camera hump on its rear panel. However, the manufacturer hasn’t given any specs, pricing or even shown us what that inside folding display will look like.
The timing of the announcement could be seen as a shot across Cupertino’s bow — Huawei beating the hotly anticipated Apple device to the punch — and perhaps that could end up being true in Huawei’s home market of China. However, Huawei doesn’t have a presence in the US, so buyers looking to broaden their foldable horizons will have to wait to see what Apple or Samsung has to show later this year.
Tech
Unitree H1 Humanoid Robot Sets Fresh Mark By Sprinting At 10 Meters Per Second

In the company’s most recent video, Unitree’s H1 humanoid bot is seen speeding down a typical athletic track and passing a speed sensor at a scorching 10 m/s (22.3 mph), with an error margin of up to 10.1 m/s. His average speed was close to the 200-meter sprint record (23.3 mph) that Usain Bolt achieved in 2009.
The company’s employees insist that although the H1 is nearly identical to human proportions and build, it can compete with the best athletes in the world. How so? The H1 weighs a comparatively trim 62 kilograms and has a total thigh and calf length of 80 cm.
Unitree G1 Humanoid Robot(No Secondary Development)
- Sleek & Durable Design: Standing at 132cm tall and weighing only approx. 35kg, the G1 is constructed with aerospace-grade aluminum alloy and carbon…
- High Flexibility & Safe Movement: Boasting 23 joint degrees of freedom (6 per leg, 5 per arm), it offers an extensive range of motion. For safety, it…
- Smart Interaction & Connectivity: Powered by an 8-core high-performance CPU and equipped with a depth camera and 3D LiDAR. It supports Wi-Fi 6 and…
To be honest, the person recording from the side appeared to be working far harder than the robot, yet the H1 just powered through the short run without even requiring any cables or support to stay upright. However, it’s important to remember that even the initial iterations of this platform were already pushing the boundaries; in 2024, for example, it was still the fastest full-sized humanoid in existence, reaching a speed of 3.3 meters per second.

High torque motors that Unitree really designed in-house and a fairly ingenious gear arrangement make up the H1’s propulsion system. Each leg contains five independent joints to offer it some degree of movement flexibility, and each arm has an additional four joints for good measure. With a depth camera and a 3D LiDAR unit that allows it to instantly adjust its footing by getting a real-time reading of the ground beneath it, this device has it all. The entire device is powered by a quite good 0.863 kilowatt hours and a 15 ampere hour battery.

All of this follows other humanoid speed tests that have recently appeared everywhere. However, a few months ago, at last year’s Humanoid Games, another machine kind of stole the show when it won the 100-meter event with a respectable time of 21.5 seconds—quite amazing. A short while later, another Chinese team reached a max speed of 10 meters per second.

Unitree’s founder stated that he expects a humanoid to race around the 100-meter track in less than ten seconds in a year. However, they will need to continue refining the balance system, strengthening the joints, and enhancing the control software in order to achieve that breakthrough. It’s not bad that the most recent run provides you a decent picture of how far these systems have come even in a very short amount of time; every little adjustment helps the robot manage its leg swings a little bit quicker without face planting or tripping over.
[Source]
Tech
Stolen Rockstar Games analytics data leaked by extortion gang
Rockstar Games has suffered a data breach linked to a recent security incident at Anodot, with the ShinyHunters extortion gang now leaking the stolen data on its data leak site.
The threat actors claim the data was taken from Snowflake environments using authentication tokens stolen during a recent Anodot security incident.
They have now published what they say is Rockstar Games data containing more than 78.6 million records.
“Your Snowflake instances metrics data was compromised thanks to Anodot.com,” reads a listing on the ShinyHunters extortion site.

Source: BleepingComputer
Rockstar Games did not respond to multiple requests for comment about the breach from BleepingComputer. However, in a statement shared with Kotaku, the company confirmed that it suffered a data breach.
“We can confirm that a limited amount of non-material company information was accessed in connection with a third-party data breach,” Rockstar told Kotaku.
“This incident has no impact on our organization or our players.”
The threat actors told BleepingComputer that the leaked data primarily consists of internal analytics used to monitor Rockstar’s online services and support tickets.
This data allegedly includes in-game revenue and purchase metrics, player behavior tracking, and game economy data for Grand Theft Auto Online and Red Dead Online. The datasets also appear to contain customer support analytics for the company’s Zendesk support instance.
In a file list shown to BleepingComputer, there were references to fraud detection systems and anti-cheat model testing.
The incident is part of a larger data theft campaign linked to a recent security incident at Anodot, a data anomaly detection company that integrates with a wide range of SaaS cloud platforms.
As first reported by BleepingComputer, the threat actors stole authentication tokens from the service and used them to access customer data stored in connected Snowflake, S3, and Amazon Kinesis instances.
Snowflake confirmed to BleepingComputer last week that it had detected unusual activity affecting a small number of customer accounts tied to a third-party integration, and responded by locking down the affected accounts and notifying customers.
The company later confirmed that the third-party integration company was Anodot.
The ShinyHunters group told BleepingComputer it was behind the attacks and claimed to have stolen data from dozens of companies using the compromised tokens.
Rockstar Games previously suffered a breach in 2022 when a hacker associated with the Lapsus$ extortion group leaked Grand Theft Auto 6 gameplay videos and source code.
Tech
What’s in Hasan Piker’s Starter Pack? Creatine, Zyns, Signal
Hasan Piker is 15 minutes late to our Zoom. He’s just gotten back from the gym, where he works out seven days a week, first thing in the morning, with no rest days. He’s been up since 5. In 45 minutes, he goes live for an eight-hour stream on Twitch. Everything is already set up, so we can chat right up until he needs to switch over, he tells me. There’s no buffer time.
Piker is a man accustomed to operating under pressure, which is fortunate because the pressure has rarely been greater. To his audience—mostly young, white men—the 34-year-old far-left commentator is a spokesperson against a failing establishment. To his critics within the Democratic Party, he is a liability protected by his “jock insurance” when he makes controversial comments about Israel and US foreign policy. Last month, a Wall Street Journal op-ed demanded that Democrats sever ties with him entirely, branding him “anti-American, anti-women, anti-Western, and antisemitic.” As Politico puts it, the left is in a Piker pickle. Piker has learned to manage with the help of his daily routine.
“Mental sanity in an insane world has to be propped up by a rigorous regimen,” Piker tells WIRED. “Think about it. Death. Destruction. An ineffective state that won’t address the working class’s needs. To maintain some semblance of hope, I have to keep my sanity, and I figured out this is the way to do it.”
I spoke with Piker—who has previously chatted with WIRED global editorial director Katie Drummond—about his relentless routine, his streaming setup, and his borderline obsession with Zyns.
When you wake up, what’s the first thing you reach for?
My phone, unfortunately. Then, my Finasteride.
Are you a coffee person?
Yes. Once I’m done tweeting, reading, and listening to NPR Morning Edition, I get out of bed and slam two double espresso shots back-to-back. And I take a bunch of pills. I take my creatine in the morning. I take fish oil pills because I throw up when I eat seafood. I take a bunch of multivitamins, ashwagandha, zinc—all that good stuff.
You stream for at least eight hours a day. When do you have the time to eat lunch?
I eat the same lunch every day, on stream, usually around 3 pm. It’s a pound of chicken. Straight white chicken breast and rice—it’s either going to be Asian or Middle Eastern chicken. I also drink a lot of cold brew while I’m slamming 3-milligram Zyns. Sometimes I substitute the 6-milligram Zyns. Coffee and cinnamon are my two flavors.
You’re often reacting to breaking news in real time for thousands of viewers. How do you avoid giving an instant take that might age badly?
It happens, but I try to be restrained. I have my ideology and message discipline on the things I’ve been talking about for years, and because the problems persist, it’s not difficult to have an instant reaction. I have talked about the necessity for gun control thousands of times at this point. So, in the aftermath of yet another horrible mass shooting, I know there are certain systemic factors at play that I can talk about instantly.
There are a lot of younger crowds watching you. How do you frame some of these political issues to them?
Donald Trump has made my job infinitely easier. My job is to educate people on imperialism and sometimes reflect on the perspective of victims. This is not an often-discussed part of our war machine. We talk about how wars impact us—our sons and daughters are sent overseas, they die, and we’re spending all of our money on it, for petro-capitalists or whatever. But rarely is there focus on the actual victims on the ground and how their perspective even shifts over time once they realize that America and Israel aren’t exactly invested in the liberation of the Iranian people as they initially presented.
Tech
What’s The Average Lifespan Of A Garbage Disposal? (According To Experts)
In North America, the garbage disposal is often seen as a kitchen essential, alongside a range of other useful kitchen cleaning gadgets. In fact, Intel Market Research reports that more than half of all newly built homes on the continent are designed to include one. It can be important, then, to understand what makes them tick, particularly regarding issues like maintenance and expected lifespan.
The International Association of Certified Home Inspectors notes that a disposal for food waste typically lasts about 12 years, approximately twice as long as that of a trash compactor (just 6 years on average) and very close to the average lifespan of a microwave. The body does note, though, that “these life expectancies have been determined through research and testing based on regular recommended maintenance and conditions of normal wear and tear.” As with all appliances, the better it’s cared for, the longer it tends to last.
Fellow professionals note that this seems to be about standard for such an appliance. Reliance Plumbing Sewer & Drainage Inc. of Northbrook, Illinois, gives a slightly lower estimate of about 10 years, while Mr. Rooter Plumbing offers a broader range of 8-15 years. Both conclude that these crucial appliances can be taken rather flippantly by users, given what they’re put through when they clear away food, and that factors such as the machine’s age and how it’s treated affect how long it will ultimately last. This is why only an average number or a broader range can be provided.
Let’s take a look at some things users can do to get the longest, most effective life from their disposal, as well as some tell-tale signs that it’s time to service or replace the unit.
Extending the life of a garbage disposal
A garbage disposal works by grinding down food waste in special chambers, its motor forcing the debris through these components via an impeller plate and out of the disposal. The more difficult we make that job, given the size and type of the waste we dispose of, the more strain our disposals are under. Only allow it to handle waste of an appropriate size, which isn’t too tough either.
Very hard waste, such as bones and nuts with tough shells, should be avoided, and congealed cooking oil or similar can harm the machine as well as your plumbing system more broadly. They are designed for softer, more manageable waste. Anything else can be disposed of in the trash, so there’s really no need to risk damaging or increasing wear and tear to the motor or the mechanism.
It’s also likely that, from time to time, you’ll inadvertently cause a blockage in your disposal. If it’s a minor issue, an equal blend of vinegar and baking soda can help break it down and clear it away, and it can also be used regularly as part of a simple maintenance routine. By making careful use of tools such as brushes with special long handles, it’s also important to directly clean the components of a disposal. The key is to monitor your disposal’s contents regularly to ensure that nothing’s building up, while not taking any risks such as using your hands or, as Molly Maid president Marla Mock puts it to Today, “us[ing] chemical drain cleaners to clear a clog, which can damage your disposal and splash back onto you.” All of these things will help to extend a machine’s life.
How to tell if there’s a problem with your garbage disposal
Of course, you can develop all the best habits when using your appliances, but nothing will stop them from aging and ultimately developing defects. It’s all about recognizing the danger signs. A garbage disposal can worsen and become more dangerous the more it’s used with an existing defect, so you have to pinpoint when it’s time to stop repairing or using it.
Liberty Homeguard provides guidance about the kinds of issues you might encounter. You might simply notice that the machine runs less efficiently, more slowly, or noisier over time. As with anything related to a home’s water supply, a leak, whether extensive or small and persistent, is a sign that something’s very wrong with the system and needs to be addressed ASAP. Smaller problems, like a jam, can be rectified by a professional or a confident user, while something more existential for the device, like its motor giving out, might mean the end.
The 10 to 12 year estimate given by this outlet for the lifespan of a garbage disposal means that, when these typical issues develop, they can be more severe and potentially more expensive, the older the system is. For instance, even a major problem is typically repairable if it’s less than five years old, according to the outlet, while with systems that have been in use for over a decade, the more practical course could be a replacement, even for a smaller problem. This ties closely to the average lifespan of the systems that other professionals suggest.
Tech
Uber and Nuro begin testing premium robotaxi service in San Francisco
If you spot a Lucid Gravity SUV blinged-out with sensors — and a self-driving system developed by Nuro — driving around San Francisco, chances are that’s an Uber employee taking a ride.
Select Uber employees can now request a ride in a Lucid robotaxi through the Uber app, the latest phase of testing ahead of a planned public launch later this year. Nuro, which provided the update in a blog posted Monday, told TechCrunch the vehicles are operating in autonomous mode and have a human safety operator behind the wheel as backup.
While this is far from a public launch, it does signal the companies’ progress since announcing a partnership and multimillion-dollar investment in July 2025. Uber invested $300 million in Lucid and separately agreed to buy “at least” 20,000 of the EV maker’s new Gravity SUV over the next six years.
Those EVs are equipped with Nuro’s autonomous vehicle system, which is powered by Nvidia’s Drive AGX Thor computer. The Lucid Gravity robotaxi, which was revealed in January, is outfitted with high-resolution cameras, solid-state lidar sensors, and radars that help the self-driving system perceive the real-world environment and operate in it.
Uber also invested an undisclosed “multi-hundred-million dollar” amount into Nuro.
The plan is for Uber to own and operate — likely with the help from a third party — the premium robotaxi service. Production of these modified Lucid Gravity vehicles is expected to begin in late 2026, according to a regulatory filing posted last year.
Nuro completed closed-course testing and started its first public road testing of the autonomous Lucid Gravity SUVs late last year. Nuro now has 100 Lucid Gravity SUVs outfitted with its self-driving system in the engineering fleet, used to gather real-world data and test autonomous driving across multiple U.S. cities and states.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
According to Nuro, the employee test rides help the team evaluate how the autonomy stack, vehicle, and rider experience work together and function in a live operating environment. It also allows the team test how well the vehicle handles rider pickups and drop-offs, a notoriously tricky operation in autonomous ride-hailing.
Tech
You Can Soon Buy a $4,370 Humanoid Robot on AliExpress
Listing consumer electronics on the internet’s large ecommerce marketplaces is a key step in “democratizing” the products, allowing them to be purchased by anyone with just a click. It has happened to cars (in the United States, you can buy a Hyundai on Amazon), and now it’s happening to humanoid robots.
The Chinese manufacturer Unitree Robotics, among the most active robot-makers in the field, is preparing to bring its most affordable model, the Unitree R1, to international markets through Alibaba Group’s marketplace. According to reports in The South China Morning Post, the rollout will initially cover North America, Japan, Singapore, and Europe. There’s no exact on-sale date for the robots yet, but the Post report says it will show up as soon as this week.
This is not the first time Unitree has used AliExpress as a global storefront. The company’s G1 model, the more powerful and more expensive predecessor to the R1, is already listed at just under $19,000.
It’s as much of a symbolic step before as a commercial one; selling a humanoid robot on a global marketplace positions the product as easily attainable. This serves as a step toward normalization of the tech, which is still not widely adopted. The sale of the R1 simply lowers the threshold of access even further, and shifts humanoid robots from the territory of promise to that of concrete availability.
Lower Price, Higher Demand
When it was announced last summer, the starting price of the R1 was 39,900 yuan, or about $5,900. Today, the basic version starts at 29,900 yuan, or about $4,370.
That price will fluctuate given changes in exchange rates and shipping costs that add on import taxes and tariffs. Still, that figure sounds surprisingly low considering that some of the R1’s other competitors in the humanoid robotics landscape are far more expensive.
The price tag for Unitree’s own flagship H1 robot approaches $90,000. Tesla’s Optimus robot, which is not yet on sale to the public, is aiming for a starting price under $20,000, but that price will only be attainable when Tesla reaches production of 1 million units a year. Meanwhile, robots from Figure AI and Apptronik are hovering around $50,000 per unit. The R1’s objectively low price essentially makes it a hatchback in a world of sedans.
The R1 is 4 feet tall, weighs 50 pounds, and has 26 smart joints. You can talk to it and give it commands; Unitree’s large-language multimodal model with voice and image recognition is on board. Curious coders can program it using a software developer’s kit. But the real calling card is the R1’s physical performance. The robot can do cartwheels, lie down and stand up independently, and run downhill. Unitree calls it “born for sport,” and videos of its presentation made the rounds months ago. Handstands and wheel kicks are not exactly what you’d expect from a robot that costs less than a used car.
Put It to Work
As impressive as the Unitree R1’s moves are, it lacks hands with articulated fingers, and its motors can’t generate a lot of torque. It is not designed to be a domestic helper or to manipulate complex objects. The company presents it as an “intelligent companion” for interaction, research, and software development.
The EDU model (Go2 EDU, G1 EDU) add an Nvidia Jetson Orin module with more computing power for artificial intelligence tasks. That model also has two degrees of freedom for the head and optional right hands. In that robot’s case, the target market is laboratories and universities. The limitations of the basic R1 put it largely in the same camp. This is not a household robot that makes coffee and walks the dog, but it is a good choice for researchers, labs, and anyone who wants to test robotics algorithms on solid hardware without spending a fortune.
It is true that bringing a relatively capable humanoid to global markets at this price does lower the barrier to entry for developers, researchers, and enthusiasts. It is a real leap from a few years ago, even if some people will buy it just to keep it in the living room to take a bow when guests arrive.
This story was originally published by WIRED Italia and translated from Italian.
-
Politics3 days agoUS brings back mandatory military draft registration
-
Sports4 days agoMan United discover Nico Schlotterbeck transfer fee as defender reaches Dortmund agreement
-
Fashion3 days agoWeekend Open Thread: Veronica Beard
-
Tech6 days agoHow Long Can You Drive With Expired Registration? What Florida Law Says
-
Politics2 days agoWorld Cup exit makes Italy enter crisis mode
-
Crypto World5 days agoCanary Capital Files SEC Registration for PEPE ETF
-
Business3 days agoTesla Model Y Tops China Auto Sales in March 2026 With 39,827 Registrations, Beating Cheaper EVs and Gas Cars
-
Politics4 days agoMalcolm In The Middle OG Turned Down ‘Buckets Of Money’ To Appear In Reboot
-
Crypto World6 days agoBitcoin recovers as US and Iran Agree a Ceasefire Deal
-
Fashion7 days agoLet’s Discuss: DEI in 2026
-
NewsBeat1 day agoPep Guardiola and Gary Neville agree over Arsenal title problem that benefits Man City
-
Crypto World11 hours agoThe SEC Conditionalises DeFi Platforms to Be Avoided for Broker Registration
-
Business4 days agoOpenAI Halts Stargate UK Data Centre Project Over Energy Costs and Copyright Row
-
Business3 days agoIreland Fuel Protests Enter Day 5 as Blockades Spark Shortages and Government Prepares Support Package
-
Politics4 days agoLBC Presenter Mocks Trump Over Iran War Failures
-
Crypto World3 days agoFederal judge blocks Arizona from bringing criminal charges against Kalshi
-
Tech4 days agoA version of Windows 10 released a decade ago is now eligible for additional security patches
-
NewsBeat2 days agoJD Vance announces ‘no agreement’ with Iran over nuclear weapons fear
-
Crypto World8 hours agoSEC Signals Exemption for Crypto Interfaces From Broker Registration
-
Business3 days agoIMF retains floor for precautionary balances at SDR 20 billion





You must be logged in to post a comment Login