Solo Founder
AI Coding
Stack
Productivity

The Solo Founder Stack: 6 Products With AI

Shihab Shahriar Antor
7 min read

TL;DR

As a solo founder I run multiple products with AI coding agents, a shared skills layer and rigorous architecture. Here is the exact stack I use to ship.

I run six products as a solo founder at Shahriar Labs — LetX, QuantumSketch, ComiKola, BikroyBuddy, BaghLang and Context-Heavy. The only way this works is a tight, opinionated stack and AI coding agents doing the boring 80%. Here it is.

The layers

LayerToolsWhy
EditorClaude Code, CursorBest-in-class code agents
Skillssoftco, skill-builder, latex-engineerReusable agent capabilities
Memorycommon-knowledgePersistent context across sessions
Inferenceopenrouter-free-inferFree LLM routing for cheap tasks
Front endReact, Next.js, TypeScript, TailwindFast, typed, low-maintenance
BackendGo (Gin), Python (FastAPI), Temporal.ioConcurrency + workflows
DataPostgreSQL, Redis, pgvectorOne database, many uses
InfraAWS, Terraform, ECS Fargate, R2IaC, no surprises

The pattern: shared skills layer

Every product reuses the same skills. softco handles "act as a complete software firm." latex-engineer handles LaTeX docs across products that need them. I covered the full set in My AI Agent Skills Stack.

Reusable skills are the only reason I can keep six products alive. Custom prompts per repo do not scale.

Front end and backend defaults

I default to React + TypeScript + Tailwind everywhere unless there is a specific reason to switch. Next.js when SEO matters; Vite when the app is primarily authenticated. Backend defaults to Go for HTTP services because the binary is 5–10 MB, GC is predictable, and concurrency is built in.

Building world-class software from Dhaka goes deeper on why footprint matters when you pay for every megabyte.

Data: one Postgres, many uses

Most products use a single PostgreSQL instance plus Redis. pgvector handles embeddings. Context-Heavy extends that pattern to a knowledge graph with recursive CTEs. One database, vector + graph + relational — covered in Building Context-Heavy.

Infra: Terraform or nothing

Every product is one terraform apply away from re-deploying. ECS Fargate for stateless services. R2 for object storage (cheap and S3-compatible). I am not running k8s for a solo shop — covered in Microservices as One Engineer.

What I don't use

  • k8s for products under 100k MAU. ECS Fargate is enough.
  • Multi-cloud. AWS or nothing.
  • Microservice frameworks. Plain Go + Gin.
  • Bespoke ORMs. sqlc for Go, SQLAlchemy core for Python.

The constraint is decision fatigue. Every "should I use X?" minute is a minute not shipping.

FAQ

Q: How do you actually keep six products running? A: Aggressive reuse — same backend defaults, same infra patterns, same agent skills layer. New product = new repo + same template.

Q: What is the single highest-leverage tool? A: Claude Code + a curated skills library. It compresses planning, coding, and testing into one loop.

Q: How do you decide what to build next? A: Distribution before novelty. If I cannot picture the first 100 users, I do not build.


Written by Shihab Shahriar Antor. Available for AI engineering and product work — hire me.

Written by

Shihab Shahriar Antor — AI Engineer & Founder of Shahriar Labs. Creator of LetX, QuantumSketch, and more.

Share this mission log