Gone but not forgotten (by at least six people)

Nobody holds a funeral for a SaaS product. There's no obituary section in TechCrunch for the apps that quietly stopped returning HTTP 200. One day they're in your bookmarks, the next day they're a parking page. The whole thing is deeply unceremonious for something that people built, invested in, and occasionally loved.

2025 was particularly brutal for AI startups. The hype bubble didn't exactly pop, but it definitely deflated, and a lot of tools that rode the initial wave of AI excitement found themselves gasping for air when the venture capital dried up and users realised they didn't actually need seventeen different AI writing assistants.

Let's pour one out.

The ones that shut down entirely

Jasper's "Jasper Art"

Okay, Jasper itself didn't die. But Jasper Art, their AI image generation feature, was quietly removed from the platform in mid-2025. The reasoning was predictable: they couldn't compete with Midjourney and DALL-E on quality, and users were overwhelmingly using dedicated image tools instead.

The lesson: being a "platform" that tries to do everything sounds great in a pitch deck. In practice, users want the best tool for each job, not a mediocre Swiss Army knife.

Character.AI's original model

Character.AI didn't technically die, but the product that made it famous absolutely did. After the Google deal, the company pivoted hard towards enterprise applications and licensing their technology. The consumer chatbot experience, the one where millions of teenagers were having conversations with fictional characters, was progressively hollowed out.

By late 2025, the free tier was so restricted that most users had moved on to alternatives. The remaining user base shrank from millions to thousands. The AI companion space got simultaneously more popular and more regulated, and Character.AI ended up stuck in a no-man's land between the two.

Mem.ai

Remember Mem? The "AI-powered note-taking app that organises itself"? It raised $23 million in 2022 on the promise that you'd never have to organise your notes again because AI would do it for you.

Turns out, people actually want to organise their own notes. The AI organisation was unpredictable, sometimes surfacing relevant notes and sometimes suggesting something you wrote three years ago about an unrelated topic. By 2025, the team had shrunk to a skeleton crew and the product was effectively in maintenance mode. The last blog post was from March. Their Twitter has been silent since June.

Lensa AI

Lensa had its moment. Remember when everyone was posting those AI portraits of themselves looking like fantasy characters? That was December 2022 and it lasted about three weeks. Three glorious, narcissistic weeks where everyone's profile picture was an oil painting of themselves as an elf warrior.

The app tried to pivot to other AI photo editing features, but the magic was gone. Monthly active users dropped by 98% from the December peak. By mid-2025, the app was still downloadable but barely maintained, a digital ghost town that exists purely because nobody bothered to take it off the App Store.

The ones that pivoted beyond recognition

Notion AI's identity crisis

Notion AI didn't die, but the version of it that launched in 2023 is unrecognisable. What started as "AI writing assistant inside your notes" has become something more like "enterprise knowledge management platform with AI features." They chased the corporate dollar (can't blame them, that's where the money is), but in doing so, they alienated the individual users and small teams who made Notion popular in the first place.

The pricing changes in 2025 were particularly unpopular. Charging per seat for AI features in a tool that many people were already paying for felt like double-dipping. The subreddit was not happy. Notion didn't seem to notice.

Replika's slow transformation

Replika went from "AI companion" to "AI wellness tool" to "enterprise emotional intelligence platform" across the span of about two years. Each pivot shed a chunk of its user base while trying to attract a new one. The romantic companion features were removed, then partially restored, then removed again. Users felt jerked around, because they were.

The ones that just became irrelevant

Some tools didn't shut down or pivot. They just stopped mattering.

AI text detectors had a rough 2025. Tools like GPTZero and Originality.ai that promised to detect AI-generated content found themselves in an impossible arms race. Every time detection improved, the AI models got better at producing undetectable text. By the end of the year, the false positive rate was high enough that schools and publishers were quietly abandoning automated detection. The tools still exist, but the confidence in their results has cratered.

AI-powered search engine wrappers flooded the market in 2024 and most of them are now ghosts. For every Perplexity (which is thriving), there were ten startups that slapped a chat interface on top of a search API and called it innovation. Most of them shut down when they realised that competing with Google's free AI Overviews required more than a nice UI and a seed round.

The "GPT wrapper" category in general took a beating. If your product was essentially "ChatGPT but for [specific thing]," you had a bad year. OpenAI kept adding features that made these wrappers redundant. Custom GPTs, while imperfect, ate a huge chunk of the market that niche AI tools used to serve.

What we can learn

The tools that survived 2025 had one thing in common: they did something that the foundation models couldn't do easily on their own. Midjourney survived because its image quality and aesthetic sense are genuinely differentiated. Cursor survived because deep IDE integration isn't something you can replicate with a chat interface. Perplexity survived because real-time search with sources is a hard technical problem.

The tools that died were mostly thin layers on top of someone else's technology. That was always going to be a precarious position, and 2025 was the year the precariousness caught up with them.

If you're building an AI tool right now, the question to ask yourself isn't "can I build this?" The question is "will this still be necessary when GPT-5 comes out?" If the answer is anything other than a confident yes, you might want to reconsider.

Rest in peace to the fallen. You were briefly useful, occasionally interesting, and almost never as transformative as your landing page claimed.