The startup stack according to AI: Gusto, HubSpot, Stripe, Clerk, Supabase, Vercel, Resend, Apollo, Amplitude, Fathom

Ten categories that make up a startup’s operating stack, and the models name a clear default in nine. Gusto, HubSpot, Stripe Billing, Clerk, Supabase, Vercel, Resend, Apollo and Fathom each lead on answer share and named-first. Product analytics is the exception: Amplitude is mentioned most and PostHog is recommended most.

Payroll

Gusto 100% share, 86% first. Rippling 98% and 8%. Deel 84% and 6%. Justworks 62%. QuickBooks Payroll 56%. OnPay 52%. All ten models agree. Gusto’s own site was the top cited vendor domain at 105 citations. Paycor and Papaya Global were never named.

CRM

HubSpot 100% and 64%. Pipedrive 100% and 16%. Attio 68% and 4%. Salesforce 62% and 2%. Close 58% and 2%. Zoho CRM 54% and 8%. Attio’s own site was the most cited vendor domain at 39 citations, above HubSpot’s 25. Salesforce is named in 31 of 50 startup CRM answers and recommended in one. Insightly was never named.

Subscription billing

Stripe Billing 100% and 72%. Chargebee 76% and 4%. Paddle 70% and 22%. Lago 58% and 0%. Metronome 48%. Recurly 44% and 0%. Paddle’s merchant-of-record positioning earns it the second most firsts. Kill Bill was never named.

Authentication

Clerk 98% and 58%. Auth0 94% and 8%. WorkOS 80% and 18%. Supabase Auth 74% and 0%. Firebase Auth 42%. AuthKit 36%. Kinde’s own site was the top cited source in the category at 99 citations, ahead of Clerk’s 31; Kinde was named in a minority of answers. All ten models most-named Clerk.

Databases

Supabase 98% and 52%. AWS RDS 92% and 24%. Neon 90% and 22%. Google Cloud SQL 66% and 0%. PlanetScale 64% and 0%. MongoDB 64% and 0%. GPT-5.6 Terra alone most-named AWS RDS; the other nine most-named Supabase. Northflank’s own site was the top cited source at 96 citations. Convex was never named.

App hosting

Vercel 92% and 62%. AWS 84% and 0%. Render 82% and 16%. Cloudflare 68% and 0%. Railway 66% and 4%. Google Cloud 60% and 2%. AWS is named in 42 answers and recommended in none, the largest single mention-to-pick gap in the stack. Claude Sonnet 5 alone most-named AWS. Reddit was the top cited source.

Transactional email

Resend 100% and 56%. Amazon SES 96% and 4%. Postmark 94% and 30%. SendGrid 86% and 2%. Mailgun 78% and 0%. Mailtrap 52% and 8%, with its own site the top cited source at 193 citations. All ten models most-named Resend. SendGrid, the long-time default, is on 43 lists and opens one. Customer.io and Courier were never named.

AI outbound and prospecting

Apollo 100% and 64%. Clay 88% and 14%. Instantly 82% and 2%. Smartlead 78% and 2%. Artisan 68% and 6%. ZoomInfo 32% and 2%. Twenty vendors named, the joint-most in the Index, and one default. Clay is the pick in 7 of 50 as the data-enrichment layer.

Product analytics

Vendor	Answer share	Named first
Amplitude	100%	30%
Mixpanel	100%	22%
PostHog	96%	48%
Heap	62%	0%
Pendo	54%	0%
FullStory	30%	0%

The models list Amplitude and Mixpanel every time and recommend PostHog most often. Open source, self-hosting and the free tier are the reasons given. Amplitude’s own site was the top cited source at 82 citations. Plausible, Statsig and Umami were never named.

AI meeting notetakers

Fathom 96% and 16%. Fireflies 94% and 8%. Granola 78% and 38%. Zoom 62% and 6%. Microsoft Copilot 60% and 2%. Otter 54% and 4%. Nineteen vendors named. Granola, the newest of the top three, has the most firsts; Fathom the most mentions. Wispr was never named.

What the field shows

The startup stack is where the models are most decisive: eight of ten categories have a leader first in 50% or more. The exceptions are analytics and meeting notes, where the most-mentioned incumbent (Amplitude, Fathom) and the most-recommended challenger (PostHog, Granola) are different products. The big-cloud options (AWS, Google Cloud SQL, Amazon SES) are on most lists and recommended almost never; the models treat them as the safe mention, not the pick.

What this means for a vendor

In the stack categories the models have a default and name it in the first sentence. A challenger’s opening is the reason the default gets replaced (PostHog’s open source, Granola’s local-first, Paddle’s merchant of record), stated in the sources the models cite for that category, which in the stack are documentation, comparison pages and Reddit.

Limitations

Fifty answers per category, ten models. Prompts assume a US-centric early-stage startup; payroll and billing answers would differ for other markets. One edition.