The AI stack we use to build SaaS companies in weeks

A venture studio has one unfair advantage: it builds the same kind of company again and again, so it can reuse what works. Our portfolio companies are small AI SaaS products, and over time we’ve settled on a stack that lets a very small team go from idea to a working product in weeks rather than months.

Tools change quickly, so this post is about the layers and how we choose within each one, more than about specific brands.

Layer 1: AI-assisted development

The biggest change in how we build is that much of the code is now written with coding agents. A founder describes the feature, the agent proposes changes across the codebase, and the founder reviews, tests and merges.

What makes this work in practice:

• Small, well-described tasks. “Add a waitlist form that sends sign-ups to our email tool with this tag” works far better than “build the marketing site”.
• Tests and previews. Every change gets a preview deployment, so review means clicking, not just reading diffs.
• The human owns the architecture. Agents are excellent at implementation; decisions about data, security and cost stay with people.

Several of our own tools, including a job-application extension and our influence dashboard, were built this way.

Layer 2: Models behind a thin layer

Products call large language models through a thin internal layer rather than directly from every feature. That layer handles prompts, retries, logging and cost tracking in one place, and makes it possible to switch models when a better or cheaper one arrives.

We match model to task: larger models for reasoning and writing, smaller ones for classification, extraction and formatting.

Layer 3: Serverless hosting

Early-stage products have uneven traffic. Serverless containers and static hosting mean we pay for usage, not for idle servers, and can scale a launch-day spike without re-architecting. Background agents run as their own services, separate from the user-facing app.

Layer 4: Content and marketing site

Every company needs a fast marketing site, a blog and basic SEO from day one. We keep the marketing site separate from the product, built as a static or mostly-static site with a CMS for posts. Pages load quickly, rank well and can be edited without touching the product.

Layer 5: Email, payments and analytics

• Email: a creator-friendly email platform for waitlists, onboarding sequences and newsletters, with tags so we know where each sign-up came from.
• Payments: a standard payment provider with subscriptions and a hosted checkout, so we never handle card details ourselves.
• Analytics: privacy-respecting product analytics plus a simple dashboard of the few numbers that matter at this stage: sign-ups, activation and retention.

What we deliberately leave out

• Custom infrastructure before there are customers.
• Microservices in the product before one service is clearly too big.
• More than one of anything. One email tool, one analytics tool, one CMS. Every extra tool is another login, bill and integration to maintain.

The principle behind the stack

Speed comes from reuse and from saying no. A small team ships faster when the boring parts are already solved and the same in every company, so all of the effort goes into the part that is actually new.

Building something?

iMelonArt Ventures backs small AI SaaS companies with this stack, our AI agents, guidance and distribution through our media brands. If that sounds useful, pitch us.