• 22 Nov, 2025 *

You might have already seen the WSJ video on the NEO 1X humanoid robot. If you missed it, watch it here, it is great.

The summary: NEO is offering an “autonomous” humanoid robot housekeeper for $20k (or a $500/month sub), and the video goes on to show how well it performs some common tasks.

Spoiler alert: not well.

Of the many notable things in the video, the item that seems to have struck a chord most with people is the teleoperation - it is not completely autonomous yet, and it is being remotely controlled by a guy in the other room wearing a VR Headset. The imagery is quite striking.

The company says that the robot is going to be tele-operated “for now” for most tasks to “collect training data”, and eventually all tasks will be automated.

The internet is having a field day with this. One of the top youtube comments on the video currently is: “It’s the self-driving car stage where the ‘self’ part is just a guy in Bangalore with a joystick”. There are also a lot of “AI = Actually Indian” jokes.

Sidebar: The person tele-operating the robot does not seem to be Indian, but this is a joke gaining steam since it was revealed that Amazon’s “Just Walk Out” auto-checkout in their Amazon Go stores was not done by AI, but by by workers in India.

Overpromising and not-delivering which seems rampant in the new AI age is also being extrapolated to be a sign of the AI bubble which we might be in right now.

There might be a ton of other reasons to suspect an AI bubble, I think it is unfair to use this as an example to support it. Though the company deserves a lot of criticism, this is part of an age-old story of product launches, nothing to do with AI itself.

Let’s dig in.

The term vapourware (skip the “u” if you are American), dates back to the 80s - a term for software which is announced but never actually released, or took way too long to release. Examples include Half Life 2: Episode Three, Duke Nukem Forever (which unfortunately abbreviates to DNF), and even Windows Vista.

Sidebar2: Vapourwave is an 80s nostalgia-bait musical genre inspired by this term which is great background music while working.

Between a Launch and a Last Place

A particularly famous case back then was an Office Suite by a startup called Ovation, which demoed its product in trade shows, but never actually delivered. It was all a ruse - the demo was faked in an attempt to raise capital. Sound familiar?

Even in the last 2010s, we saw several EV companies (Faraday Future, Nikola) - promise the world, and disappear. Magic Leap is another example, we never got to see their AR glasses.

This phenomenon is not new, and not limited to AI. Seems to be common in any currently “hot” field.

The issue here is that the companies are trying to prioritising other goals - raising capital, increasing stock price etc., over the actual promise to the customer. In some cases it is about being first to announce something and put it in the market.

I sympathise with this. Startups, especially consumer hardware startups, are a tough business. You have to show “traction” with customers and get some media hype to raise capital to cover development costs. Once you get that traction, you need even more funding to get to production, scale and build distribution.

And sometimes, companies don’t make it through this grind. Its sad.

I also have a bit of a bias, since I worked in hardware companies which launched early - and made it. Ather launched the S340, India’s first intelligent electric two-wheeler, at an event in 2016, while the first scooters were delivered only in late 2018. We also talked about a ton of software features - which would be delivered in the future as over-the-air updates.

So, is it bad to announce the product before it is sufficiently ready?

No, if you only announcing it. Get people excited about it. Get some eyeballs. Acquire some leads. Raise some capital. Hope to see you guys on the market!

Signing up for updates on the website costs users nothing, and VCs have teams of analysts doing due diligence before funding any company, so its all good.

Yes it’s bad, if you are taking my money for it. If you are listing your product for pre-order, you should be at least be in the last stages of development, about to move to the production phase. Ather opened opened pre-orders only in mid-2018.

Depending on your product this part may take 1-6 months. Which is fine. But charging $50,000 and making people wait for 8 years is bad, whether it eventually has a rocket engine or not.

NEO’s pre-order price is $200, while promising a delivery vaguely “in 2026” - I am putting it in my “bad” bucket.

Over-the-air-everything

With the rise of connected devices, it is now common for hardware product launches to promise some software-based features in the future.

There are several reasons to do this.

Reducing time to market being the main one. While building hardware + software products, software development on a new platform happens AFTER the certain stage of hardware development. Which means you might have to delay the launch of the product while you develop the software, though your hardware is ready.

Solution? Build core features, ship and update the rest on the fly.

I don’t mind this at all. Customer gets the product quickly, with the necessary features. And the product gets better over time.

However, some products miss the part about building the “core features” and ship a hardware that is not very useful.

So much so that popular Youtuber MKBHD tweeted this.

I agree with this. When you are buying a product, buy it for the features it has currently. Whatever you get later, is icing on the cake.

If the product is defined by a feature, then it has to have it at launch. The nature of the product could evolve - a smartphone today is defined by its apps. But when it was launched it was, very famously, a phone + an iPod + an internet navigator. Whether it does the feature well or not will define the product. And hopefully you survive long enough to see the product evolve.

The second reason to update the feature later is because some features need data to exist. YouTube could not have launched with recommendations on the first day. Google Maps’s first version had only the US and British Isles, and they could not have promised navigation with live traffic data.

So if your core feature is dependent on data, you need to find a different initial pitch. For example, main pitch: a health tracker monitors your vitals, but over time, it provides personalized insights, which makes it more interesting and indispensable. If insights is your day 1 pitch AND a differentiator from existing products, I expect you to have the algorithm ready at launch.

Looking at this lens, building autonomy is a data problem.

Tesla’s FSD package has been available for almost 10 years now, but the cars are still learning, only moving from “beta” to “supervised” last year.

NEO has the same issue.

The “foundation models” of robotics - which are expected to be the brain of these humanoids - are lagging behind compared to LLMs, mostly for the lack of training data. There is an internet worth of text and image data for LLMs/VLMs, but barely any robotics data out there for the VLA (Vision-Language-Action) models being built.

So, teleoperation is seen as one of the primary ways to operate robots and gather these data.

Despite the context, the primary feature of your robot is that it is an autonomous robot. It is not an autonomous robot today. I appreciate the fact that you are upfront with the tele-operation unlike some others in the field, but that is not enough. Sell a tele-operated robot service today. Change the pitch to autonomous robot once it is ready.

Has this been done before?

Features should be promised only if there is reasonable evidence of achieving them.

This means:

• You have built a POC of this feature (or variations of it) and tested it to a reasonable degree, and are unable to launch it now. Sometimes features take time to productionize, so you launch the core product now with the promise of another feature in the future. That’s ok.
• The feature has been live in the field in other contexts, so the success of your implementation of it is an eventuality.

At upliance.ai, we built smart cooking machines. Though we had a mobile app connected to our cooking machine, it only showed the recipes available on the device, and did not have a “cooking done” notification, which the users would find useful. We promised this feature in a future update, as there was no actual technical challenge in building this feature.

At Ather, we promised that the Ride Statistics on the electric scooter’s companion app would be available in a future update. Though we had not built out the stats internally, we had used this type of data for our internal testing. So we had 100% confidence of pulling off the feature post the launch. Ather was/is a personification of under-promise and over-deliver.

But - there have been some lapses, which could have done better. At upliance we promised a “grocery list” with quick commerce apps like blinkit or instamart so users can easily add all the ingredients needed for their recipes to cart easily on their favourite apps. There is no technical challenge to achieve this. However, it is dependent on the third-party quick-commerce API being available.

Doordash in the US seems to provide similar APIs, but the Indian grocery delivery companies don’t - and don’t seem to have it in their immediate roadmap.

Ather released a pothole detection feature recently, though we first talked about this feature back in 2017. This was a data-driven feature, so it needed a lot of scooter data on the road before starting to identify potholes.

But, there was another reason - priority.

As a startup, you have millions of things to do, and you will prioritize features/bug fixes which your (hopefully) customers need the most. This might mean that some features which are possible but not necessary get deprioritized until you have the bandwidth to get to them.

So though it is understandable that the startup does this, if you are a customer who is buying the product for that particular feature, you are out of luck.

Repeat with me: do not buy the product for promise of future updates, buy it for what it is today. For well intentioned startups: do your best to prioritize the features your customers are asking for.

Hardware Is Ready

Tesla-FSD-Beta-pedestrian-Blue-Ridge-Mountains

Tesla’s promise was that all the cars with the FSD hardware will have FSD. There have been multiple iterations of the FSD hardware - and there will continue to be as the problem and the solutions to the problem - hardware, algorithm evolves and throws up new requirements.

When you are dealing with a new wave of technologies, it is hard to predict - and as humans we tend to be overly optimistic about it.

MagicLeap needed breakthroughs in the lenses, processing, batteries, image processing and AI to make their glasses work.

There are a lot of variables when you try to pull off a project of this ambition - you are betting on the hardware, the software, the configuration of the sensors, the regulation, the compute, the cost - all working well to deliver your promise. So if the stars don’t align, you are NGMI.

If you showing a novel product to the world, especially in the intersection of hardware and software, you will not know what you can deliver until you test the whole stack.

Even Apple, who is usually really good at delivering on promises, announced Apple Intelligence on its iPhone 15+ hardware, but has not delivered it as of 2025.

Same is the issue with NEO 1X - the core algorithms needed to make the robots autonomous today are not a solved problem. There have been interesting developments in VLA Models and World Models but they are nowhere close to full autonomy. Though the company says the Robot has all hardware needed for the future models work, it is not a sureity. It might need an upgrade in the camera resolution or processing power to use the models when they are ready.

Promising that this particular hardware is ready for deploying an algorithm and enabling features you have not tested yet is naive at best, and is most likely intentionally misleading. With NEO, I am tending towards the intentional misleading part.

Final Thoughts

Summarizing my thoughts on this topic from two perspectives. First as a early stage startup, and second as a consumer.

Your startup needs to make noise and garner interest both for early customer traction and for investor interest. So go ahead and paint a vision of future where your product rocks the customer’s world. Once you start asking for money for the product, things change. Take pre-orders when you are confident of hitting the market in the short term, and have actual confidence of going into production. Deliver the core features which define your product. It is ok to promise cool updates in the future, as long as they are not the core features and there is minimal technical challenge in implementing them. This ensures that you are able to actually able to deliver what you promised. Lastly, listen to your customers, and build things people want.

As a customer, buy the product only if it gives you what you want on day one. If you trust the company, go ahead and purchase it for the future features, but keep in mind that it might not happen for various reasons.

However, that does not mean you let companies off the hook for actually not delivering on promises. Email them, post it on their community and social media - show them you care for it. Might actually change their development priorities.