_SH Log's
← Back to Root
EST: 4 min read

The Solo Founder Stack: 6 Products With AI

As a solo founder I run multiple products using AI coding agents, a shared skills layer, and rigorous architecture. Here's the exact stack I use in 2026.

#founder#ai#productivity#systems

As a solo founder running six live products — LetX, QuantumSketch, BikroyBuddy, Bagh, ComiKola, and offSchool — the only reason I'm not buried is a disciplined AI-assisted stack. Here's exactly what I use and why each piece earns its place.

The three-layer model

Layer 1: AI coding agents (Claude Code + custom skills)
Layer 2: Shared infrastructure (AWS, Terraform, Temporal)
Layer 3: Per-product code (Go services, Next.js frontends)

No layer bleeds into another. AI agents operate at Layer 1 only — they read and write code, not infrastructure. Terraform at Layer 2 means no product-specific AWS console clicking.

Layer 1: AI coding agents

I use Claude Code as my primary coding agent with a custom skills layer I built and open-sourced:

| Skill | What it does | |-------|-------------| | softco | Full product lifecycle — PM, architect, senior dev, DevOps | | skill-builder | Builds new agent skills from plain-English description | | frontend-design | Production-grade UI avoiding generic AI aesthetics | | seo-master | SEO + GEO + AEO for all blogs and landing pages | | common-knowledge | Git-backed persistent memory across sessions |

The key discipline: AI agents write code, I review architecture. I never let an agent make infrastructure decisions (which AWS region, which DB engine, scaling strategy). Those stay in my head and Terraform.

Layer 2: Shared infrastructure

Every product shares the same AWS account and Terraform mono-repo:

# modules/service/main.tf — every product uses this
module "letx_api" {
  source        = "../../modules/service"
  name          = "letx-api"
  image         = var.letx_api_image
  cpu           = 512
  memory        = 1024
  port          = 8080
  env_vars      = local.letx_env
}

Shared modules: ECS Fargate cluster, RDS PostgreSQL (multi-tenant), S3 buckets, CloudFront distributions, Route53, ACM certificates. Adding a new product means writing ~30 lines of HCL, not a week of DevOps.

Layer 3: Per-product code

Every product follows the same tech-stack template:

  • Backend: Go + Chi router + sqlc + pgx
  • Frontend: Next.js 15 App Router + TypeScript + Tailwind
  • DB: PostgreSQL (shared RDS, schema-per-product)
  • Queue/workflows: Temporal.io for anything async > 5 seconds
  • Auth: Custom JWT (Go middleware, shared library)

Copy-paste is fine here. Premature abstraction across products is not.

How a typical feature ships

  1. Open Claude Code: /softco build [feature] for [product]
  2. Agent generates: spec → architecture → code → tests → Terraform delta
  3. I review architecture and DB schema changes (never skip this)
  4. git push → GitHub Actions → ECR push → terraform apply
  5. Done

A feature that would take 3 days solo takes 4–6 hours with this loop. The bottleneck is always my review, not the agent.

What breaks this model

Three failure modes I've hit:

  1. Agent accumulates bad context — long sessions drift. Fix: /clear + re-scope the task.
  2. Shared infra conflicts — two products needing the same RDS instance size. Fix: separate RDS per product once revenue justifies it.
  3. AI-generated tests that pass but don't test anything — always read the test, not just the green checkmark.

Cost (all 6 products)

| Item | Monthly | |------|---------| | AWS (ECS + RDS + S3 + CF) | ~$140 | | Claude API | ~$35 | | ElevenLabs (QuantumSketch) | ~$22 | | GitHub Actions | $0 (free tier) | | Domains + misc | ~$15 | | Total | ~$212/mo |

Six live products for $212/month is the leverage point. Revenue from LetX and QuantumSketch subscriptions covers everything with margin.

FAQ

How can one person run six products? Aggressive AI leverage, shared infrastructure, and strict scope discipline. Each product only gets features that move revenue — no nice-to-haves.

What AI coding agent do you use? Claude Code with custom open-source skills for product management, architecture, frontend design, and SEO.

Is Go or Python better for AI backends? Go for APIs and orchestration (fast, low memory, easy concurrency). Python only where a library is truly irreplaceable (Manim, some ML inference).

How do you avoid context bleed between products? Each product lives in its own Git repo. Claude Code opens one repo at a time. Shared code is a library, not a monorepo.

Written by Shihab Shahriar Antor — AI Engineer & Founder of Shahriar Labs. See also: My AI Agent Skills Stack · How I Use Claude Code to Ship Faster.