From October 2022 to October 2023 I ran a mental health startup out of Pune called MindWave. Thirteen months. The core product,private sharing, journaling, and venting,shipped. Real users came. Real users came back. We were sketching out group therapy and a free community for mental health professionals as the next two layers when I shut it down. This post is the postmortem I wish someone had written for me before bet number one.

It belongs to a larger pattern. By early 2026, 40% of AI startups launched in 2024 had already shut down (TechStartups, December 2025). I'd already shut down three startups before that wave hit, and MindWave was attempt one. The structural failure here is not an "AI" failure,it's a "marginal improvement in a high-emotional-switching-cost category" failure. That category is brutal whether or not your product has an LLM in it.

What MindWave was

MindWave started as a private sharing product. Three sentences capture the thesis: people need a place to vent that does not become a permanent record on a social platform; that place should feel emotionally safe; and the existing apps treated mental health like a meditation playlist when the people I cared about needed something more direct.

The product was a private sharing feed plus journaling, with anonymity controls and a tone calibrated for late-evening use. The founding location was Pune. Beyond the core, two planned layers: structured group therapy sessions, and a free community for mental health professionals so that supply and demand sat on the same platform. The professional community was the part I was most excited about. It did not ship before shutdown.

What worked

Three things worked, and they're worth naming because the failure mode here is not "no traction". It's "real traction that does not compound to escape velocity".

First, the right user showed up. People who tried existing meditation apps and felt under-served by their evergreen-positivity tone found MindWave on their own. The acquisition was largely organic. The session content was real,long messages, repeat visits, real expression. The kind of usage you cannot fake.

Second, professional interest validated the demand-side thesis. The conversations I had with practising therapists confirmed that the supply side existed and would have plugged into a free community layer if we'd gotten there. The supply-side moat was real.

Third, the team was emotionally aligned. Building a mental health product is harder than building a productivity product because the responsibility is heavier. The team treated that weight seriously. That is rare and worth saying out loud.

What didn't work

The core problem was simple to state and hard to fix: every time I described the product, I was selling on care and warmth, not on a measurable improvement over an existing alternative. "It's gentler than what's out there" is a feature; it is not a category-shifter.

The user's question was never "is MindWave exactly right for me?" It was "is MindWave so much better than what I already use that it's worth switching?" The honest answer was no. Not in a way that mattered. People who already had a relationship with BetterHelp, Wysa, or Calm did not abandon them; people who had not yet entered the category often used the free, well-marketed incumbent's onboarding because that's where the funnel pointed. We were a slightly better experience inside a hard-to-reach pocket of the market.

"Better than incumbents" needs a much bigger gap than founders think Founder's perceived gap ~1.4x User-perceived gap ~0.3x Required to displace 10x Even a real, measurable improvement does not move users in high-emotional-switching-cost categories without a category-shift. Source: Aryan Agarwal, MindWave operator review, October 2023. Gap framework: Peter Thiel, Zero to One.
The "10x rule" gap. MindWave's actual user-perceived gap was a fraction of what was required to displace.

The second issue was sequencing. The professional community would have been the unique layer,supply-side gravity that incumbents structurally could not copy. We were trying to ship the consumer side first and then layer the professional community on top. That sequence is wrong for a marketplace dynamic. Marketplace products that win usually solve the supply side first and pull demand toward it. We were doing the inverse.

The third issue was that I underestimated how much of the value was emotional context, not feature work. The product was easy to clone superficially and hard to clone in tone,but tone is not a moat that compounds against incumbents who have years of brand work behind them. We had no compounding mechanism that incumbents could not match by hiring two more people and rewriting their copy.

The shutdown decision

The decision moment was not a single bad metric. It was a description.

I was on a call in late September 2023, walking a friend who'd asked through "what does MindWave do?" Halfway through the third sentence, I noticed I had used the words "gentler" and "more caring" and "actually listens" and not a single phrase that pointed at a measurable improvement. There was no "10x faster", no "10x cheaper", no "10x more accurate", no "10x lower friction". The pitch had a tone, not a metric.

That call ended. I closed the laptop and ran the numbers I'd been avoiding,month-over-month organic acquisition, the curve from week-three retention onward, the conversion from free user to professional connection. They were fine. Not catastrophic. But "fine" was the answer for a product the user thought of as "fine". Fine is not a startup outcome. I shut MindWave down two weeks later.

The honest hindsight read is that I should have shut it down four to six months earlier. The signals were there in month six. I rationalised through them because the team was good and the users we did have were the right users. Both true. Neither sufficient.

Why mental health is structurally harder than it looks

Mental health is a high-emotional-switching-cost category. The user has invested trust in their existing tool,even if that tool is mediocre,and the cost of switching is not just signing up; it's re-explaining context, re-trusting, re-disclosing. The bar to displace is structurally higher than in productivity software, where switching costs are largely about data export and a learning curve.

The category also has the highest safety bar in consumer software. Yara AI's founder shut down her own AI mental health companion in November 2025 because, in her own words, AI chat for serious mental health was too dangerous (Fortune, November 2025). That's a founder choosing to shut down a working product on principle. It speaks to a category dynamic that founders entering this space need to model in: the floor for "responsible" is much higher than the floor for "shippable", and the gap between them is large enough to swallow a year of runway.

Across the broader 2025-2026 AI shutdown wave, mental-health-adjacent products consistently came up in the failure roundups (TechStartups, December 2025). Not because mental health is intrinsically a bad market,but because the bar in the category is much higher than most founders model in.

The lesson,and where it lives now

The lesson tag for MindWave's failure: massive-value test failed. Real engagement, real users, no 10x. That maps to one of the three checks in the synthesis framework,see the full three-startups synthesis for the full framework, plus how Super AI failed the scaling test and Vibe AI failed the margin test.

For Gravity, the MindWave lesson surfaces in the deploy-time bet. Sixty-second deploy is a measurable, falsifiable claim,not "easier than Zapier" or "more autonomous than ChatGPT". That sentence is the kind of pitch I could not have written for MindWave. It's the test I run on every product decision now: can I name the dimension and the multiple? If the answer is "kind of, on tone",that's the same answer MindWave had. It is the wrong answer.

If you're building in mental health right now and want to compare notes,or you're evaluating whether your product crosses the 10x threshold,I read every email at aryan@gravity.fast. The About page has the company background.

Frequently asked questions

Why do mental health startups fail?

Mental health startups face a higher "better" bar than most consumer categories: incumbents like BetterHelp, Calm, and Wysa already deliver real value, and switching costs are emotional, not just functional. A marginal improvement is invisible to users. MindWave's core product was useful but not 10x better, and that gap killed it within 13 months.

What was MindWave?

MindWave was my first startup, founded in October 2022 in Pune. A mental health platform centred on private sharing, journaling, and venting, with planned layers for group therapy and a free community of mental health professionals. We shipped the core sharing product, saw real engagement, and shut down in October 2023.

How long did MindWave run before shutting down?

About 13 months,October 2022 to October 2023. The product worked in the sense that real users showed up and used it consistently. The shutdown decision was driven by recognising that "gentler than the alternatives" was not a 10x improvement on any measurable dimension.

Are AI mental health apps safe?

AI-driven mental health apps face safety questions that the founders themselves often raise. Yara AI was shut down by its own founder in November 2025 because she concluded AI chat for serious mental health was too dangerous. Mental health is one of the highest-stakes consumer categories, and the safety floor,not just the product floor,is structurally higher than most AI verticals.

What is the lesson from MindWave's shutdown?

"Gentler" is a feature, not a category-shift. In a market with strong incumbents and high emotional switching costs, the bar to displace is much higher than the bar to be useful. MindWave was useful. It was not 10x. That gap is sufficient cause of death even when retention metrics look reasonable on a small base.

Three takeaways before you close this tab

Sources