On 14 November 2026, TechCrunch published a running list of AI projects and startups that shut down or missed expectations. It covers Character.ai, which filed for bankruptcy in early October 2026. It covers Relay, a workflow automation tool that closed in August 2026, and Ollie, a family AI assistant scheduled to shut down on 6 November 2026. It also covers bigger names. Apple’s improved Siri was announced in June 2024, delayed repeatedly, and only began rolling out to English-language users in November 2026. OpenAI’s “super app” update on 9 July 2026 combined Chat, Codex and Work modes. Users found it confusing and cluttered, and OpenAI rolled it back soon after.

The excitement was about what the tech could do

Each of these launches came with a story about what AI could now do. A Siri that understands context and acts across apps. One app for chatting, coding and working. A workflow tool as an alternative to Zapier. A podcast-style audio app. An AI email product.

Those are impressive capabilities. I separate “technically impressive” from “where this is actually going”, and the list shows why. A capability is not a customer with a problem.

The same pattern shows up in the failures

Read the reasons the article gives and a pattern shows up.

Relay shut down because larger platforms like OpenAI and Google built similar automation into their own tools. That left it without a clear reason to exist on its own. Huxe closed partly because Spotify began offering similar features to a huge existing audience. Notion found that users were running separate AI agents to manage their inboxes, so Notion Mail is scheduled to shut down on 22 September 2026. Ollie’s CEO said it failed to stand out against rivals like Muse and Instinct.

In each case, the person using the product already had a way to get the job done. The new product had to beat that, and it couldn’t.

The OpenAI example is a slightly different version. People already had a ChatGPT they understood. Adding modes and renaming the old one “ChatGPT Classic” gave them more to learn, and the stated problem it solved was hard to find. Users pushed back and the old interface came back.

My take: the first question is always whether there is a need

When a new AI product launches, I don’t ask what it can do. I ask what problem it solves that I couldn’t already solve, and whether that is worth adding to my day.

Take a clinic owner who sees a new AI scheduling assistant, or a retailer who sees a new AI inbox. They are told it will save hours. The useful move is to write one sentence first. “This solves X, and my current tool cannot.” If the sentence turns into “it does the same thing with a nicer interface”, there is nothing to adopt. If the sentence can’t be written at all, that is the answer too.

The second half matters just as much. Fewer tools, used deeply, beat many tools used shallowly. Every new tool costs setup time, learning time and another place where your data and habits get split. I use Claude for almost everything, and I find one tool used well more efficient and more effective than a stack of them. A team that knows one assistant deeply gets better results from it each month, because they know how to brief it and where to check its work. A team that tries six tools for a week each gets the first-day experience of all six.

The graveyard also shows what happens when the bigger tool absorbs the feature. Relay, Huxe and Notion Mail all lost ground to something people already had open. If you go deep on a tool that keeps adding what you need, you are less likely to be stranded when a standalone product closes.

Where this touches what I build

I think about this with AeroChat too. Every AI chatbot is almost the same at the chat level, so if AI is the only selling point, it is not a selling point. The question for a business owner is whether their customers’ requests, like “where’s my order”, turn into actions in their own systems. That is a need someone already has, and it is the test I apply to AeroChat as well.

What a business should do

Before you adopt any new AI product, write down the problem it solves that your current tool does not. Keep it to one sentence. If you can’t write it, skip the product.

If you can, test it on one real task from your normal week, not on the demo. Compare the result with what your current tool gives you on the same task.

Then pick the one tool you already use most and set a goal to use it more deeply. Learn how to brief it properly and write down what you check in its output. Add a second tool only when the first one clearly can’t do a job you need done.

Source: TechCrunch: The AI graveyard: a running list of projects and startups that didn’t make it