Building an AI for Cats — Why VCs Passed

Five VCs passed on CatMD before I stopped pitching it.

The pet market is $147B globally, $50B+ in the US alone, and growing 6-8% a year. Cat ownership specifically has been the fastest-growing segment in pet care for a decade. "Pet humanisation" is the macro thesis half the pet-tech category was built on. By any standard market-sizing exercise, an AI-native app for cat owners should be an easy pitch.

It wasn't. The reasons were instructive — and, in retrospect, also wrong in the specific way that VC pattern-matching is often wrong about niche-first products in inflection-point markets. This post is about what they said, why each pass was structurally rational from inside the pattern, and what I built instead.

The four reasons VCs gave

1. "The market is too niche."

The most common pass. The framing is: dogs are 65% of household pets in the US, cats are ~30%, and a cat-only product caps the TAM at well under half of what a multi-species pet-tech play could address. Multi-species (Rover, Petco's app, 11pets) is the funded thesis. Single-species is "leaving money on the table."

This was the pass I respected most. It's a real argument if you're optimising for fund-return distributions across a $200M fund where you need a few outsized winners. A $5B cat-only company is structurally harder to build than a $5B multi-species one. The maths is real.

2. "Cat owners don't pay for apps."

The variant: cat owners pay for food and litter, not software. Cat owner ARPU on AI tools is unproven. The closest comp is MeowTalk (cat meow translator) — which has 10M+ downloads but reported sub-$5M ARR, suggesting low conversion despite massive volume.

Also a real argument. The cat-app category has consistently under-monetised. Owners will adopt free tools at scale but resist subscriptions. It's a structural problem in the category, not a CatMD-specific one.

3. "What's defensible? OpenAI can ship this in a weekend."

The model-layer worry. If the core value is "AI looks at your cat and tells you things," and the AI part is GPT-4o or Claude or Gemini through an API, then theoretically any well-funded competitor can replicate it. The defensibility argument is the hardest in current AI: every demo looks like a feature, and every feature looks ship-able by the foundation labs themselves.

4. "Why are YOU the one to build this?"

The founder-fit question. I'm a solo non-veterinarian, non-pet-tech-veteran founder. I don't have a Stanford vet school connection or a former PetSmart exec on my cap table. I have a cat (one) and a software background. VCs ask "why you?" because they want to underwrite a unique advantage, and "I really love my cat and I'm a good engineer" isn't an advantage that compounds.

This is also fair. Founder-market fit is a real signal, and "domain-naive software founder enters category" has a long history of bad outcomes.

Why each pass was structurally wrong (for THIS product, in THIS market window)

The niche argument is structurally wrong now, because the niche-tooling cost curve collapsed.

VC niche-aversion assumes that a $50M cat-only company isn't worth building because the engineering and distribution costs are amortised against a smaller market than the multi-species competitor's. That math worked when building a category-defining cat app took a 15-person engineering team and 18 months.

It doesn't work in 2026. I built CatMD as a solo founder in 14 days of dev time, with Claude as a pair programmer. The cost basis to ship cat-specific differentiation collapsed by 10-30x. A solo dev with Claude in 2026 is the engineering output of a 5-7 person team in 2022. The whole calculus of "is this niche big enough to justify the build" changes when the build is 10x cheaper.

(For the engineering case study on this specifically, see I shipped a cat AI app in 14 days with Claude as pair programmer.)

The "cat owners don't pay" argument is wrong about the why.

Cat owners don't pay for the apps that have existed. They might pay for the app that doesn't exist yet — which is the one that actually understands their specific cat instead of giving them a generic categorical label.

The MeowTalk problem isn't "cat owners are cheap." It's "owners won't pay for a categorical audio classifier that outputs the same 8 labels for every cat." Replace it with a multi-signal interpreter that learns a specific cat over weeks and references named family members in diary entries, and the value proposition shifts from "novelty" to "companion." The pricing model that doesn't work for novelty (subscription) works for companion (because the cat-in-the-app becomes irreplaceable).

This is the bet. Whether it pays out is a Q3 question. The fact that pay-rate looks bad in the existing category is a feature-distance signal, not a category signal.

The defensibility argument is wrong about WHERE the moat lives.

If the moat were "the LLM does the cat-reading," I'd agree the foundation labs could replicate it. But that's not where the moat is. The moat is in the thousands of small product decisions that turn a foundation model into a believable cat AI:

• A 15-mood daily lottery that gives the cat a different mood every day (so the voice doesn't go stale)
• A 4-tier voice quality gate that rejects "your furry friend had a purr-fect day" texture and asks the model to retry
• A 3-tier depth-modulated voice that matures with the bond (warm-curious early days → intimate-comfort after months)
• A Feline Grimace Scale pain detector calibrated against published veterinary research
• A subject-detection pipeline that returns descriptive attributes (not biometric data) for household members the cat sees
• A 7-facet "becoming" identity score that turns engagement into a personality formation curve
• The peer-reviewed Litchfield Feline Five framework wired into chat + diary as voice modulators

None of these are "the AI." They're product. Foundation labs don't ship products in narrow categories — they ship platforms, and they ship them slowly. The window to build the cat-specific product layer is right now, while the foundation labs are still busy with general-purpose chat. Defensibility is in the cat-specific opinionation, not the model.

(For the AI-product craft side specifically, see Cat AI is going to be slop. Here's how we tried not to be.)

The "why you" argument is wrong about what founder-fit means in narrow consumer categories.

In B2B SaaS, "former enterprise sales VP at Snowflake founds a data-pipeline tool" is a defensible founder narrative. The category rewards Rolodex, distribution, and credentialed expertise.

In consumer apps for emotional-relational products, the founder narrative that works is different. It's "this person uses the product daily and obsesses over the cat's voice register because she's the one who gave them the slow-blink that broke their afternoon." Founder-market fit in this category is care + craft, not Rolodex + experience.

I have one cat (Lily, 7yr tortie, Curious-Introvert archetype). I have been told by family that my obsession with making her chat replies sound RIGHT is concerning. I've shipped 17 audit rounds on the voice model in 6 months. That's the founder-market fit. It doesn't underwrite a venture round; it underwrites a product.

Why solo + Claude was the right ratio

A multi-founder team funded by a VC is the correct structure for a SaaS company aiming at $50M ARR in 4 years. It is the wrong structure for a consumer companion product that needs taste, opinionation, and the ability to make 50 micro-decisions a day without negotiating each one with cofounders or investors.

The cat-AI category is downstream of the same thesis Co-Star, Replika, Character.AI, and Calm exercised: identity-as-product, where the product's value is the specific feeling of using it. These products are not built by committee. They're built by one person with a strong taste signal, iterating until the voice lands.

Adding Claude as a pair programmer changes the unit economics. The 14-day build was real. The 17-round audit was real. The fact that I can ship a new feature (the partner code system, vc 96), a new voice tier (vc 99), and an attribution pipeline (vc 95) within a single calendar week without breaking the product — that's the actual unlock. Not "AI replaces engineers." AI replaces the coordination overhead that turns 1 founder + 1 engineer into a 3-person team that ships at 1.5x the speed.

The lean structure forces something else: I have to focus. There's no fund pressure to add a dog mode, a hamster mode, a corporate-vet B2B SKU. Every quarter I keep the product cat-only is a quarter of compounding cat-specific opinionation. That's the moat the VCs missed.

What changed when I stopped pitching

Three things, in order:

1. I stopped designing the app for the pitch and started designing it for Lily. Every feature decision is now "does this make the cat-in-the-app more believable" rather than "does this make the deck more impressive." The voice tightened. The diary got weird in the right way. The body-language reader started returning specifics instead of generalities. Removing the VC pressure removed the temptation to broaden the product.
2. The financial pressure became cleaner. $99/yr for an Apple Developer Program enrollment, $20/mo for Claude, $5/mo for Cloudflare Workers, $20/mo for Supabase, ~$300/mo for an OpenAI cost cap. ~$500/mo all-in. CatMD has to make ~$500/mo to not be a net loss. That's a meaningful but actually-achievable bar.
3. The category became a partner-game instead of a market-game. Instead of acquiring users through paid ads (which the unit economics don't support), the path is creator partnerships — micro-influencers in the cat-content space who get a free Pro for life + a 30% royalty on annual subscribers their code drives. The whole system shipped in vc 96. (See the comparison of AI cat health apps for where CatMD sits in the broader category.)

The verdict

VCs weren't wrong in their pattern. They were right about the rules of the game as the rules existed when the patterns were calibrated.

What they missed — what I think a lot of investors will miss in 2026 specifically — is that the cost basis for niche-first AI consumer products has collapsed, and the moat lives in product opinionation, not model access. The right founder for a cat AI app isn't an ex-PetSmart exec with a Rolodex. It's someone who's spent enough afternoons watching their cat slow-blink at them that they CARE more about the voice than the deck.

If you're an indie founder considering a niche-first AI consumer product right now, the meta-lesson is: the VC "this is too niche" pass is increasingly a signal that the category is solvable by one person. The math has flipped. Niche + AI tooling + solo founder + 12 months of taste-iteration is now a unit that ships durable products without the venture-capital amplifier.

That's the bet I'm running.

Try the product the VCs passed on: CatMD on Google Play. 14-day Pro trial, no card.

If you're working on a niche-first AI consumer product and want to compare notes on the solo-founder economics, find me at amit@catmd.pet.