CTO Partner — BOSS-IQ

Who is writing this

• I am a business-development and marketing strategist — 30+ years of it.
• Up until December 2022, when ChatGPT first came out, I had no technical background at all. Since then: obsession. Constant research, self-taught AI engineering, and small forward-deployed LLM systems built for myself and for clients.
• Over the last 3 years, AI-transformation consultancy and corporate prompting workshops grew from a side offering inside my strategy practice into the focus of it.

How BOSS-IQ got here

• 14 months ago I started building BOSS-IQ, confident I could vibe-code the whole thing alone.
• I wrote the prompts first — essentially the same prompts running the product today: 18,500+ lines across 79 agent prompt files. The codebase came after, to serve them.
• Through the summer of 2025 I tried to build that codebase myself, in Cursor. When I realized I could not get out of the spaghetti code I had created, I did what a strategist does:
  • wrote the business plan,
  • raised a friends-and-family round in December,
  • hired a professional contract developer.
• December to February: he built the code infrastructure of the architecture I had designed — roughly the first 20,000 lines — and delivered a basic working product.
• Since February, I have been the only developer. Under my prompts, the codebase has grown more than tenfold — about 264,000 lines of TypeScript today.

The product, today

• Hand-onboarding our first paying customers since mid-August, invitation-only.
• 60+ specialist AI agents maintain each customer company's Intelligence Library — patent-pending since July.

The development environment — the technical part

• What makes solo vibe coding survivable at this scale: while coding freely in Cursor, a semi-autonomous rule network oversees everything. Even the coding is prompting — I author intent; the environment produces, checks and documents the code.
• 45 interconnected instruction files — 300,000+ words of standing rules — orchestrated by 1 always-loaded kernel. Every session, it:
  • classifies the work,
  • auto-attaches the domain files matched to the exact code being touched,
  • fires protocols on trigger phrases,
  • plans the work into a delegation topology and dispatches 8 sub-agents across 3 model tiers — autonomous code reviews, session audits and mechanical edits run as sub-agentic processes, planned at the kernel level,
  • captures every transcript, decision and code review back into the rules.
• The system is self-documenting — journals, review records and domain knowledge are written by the system itself at session end. Documentation has been automated from the start, not deferred.
• Nothing depends on my memory. The rules remember, the rules check, and the rules teach the next session what this one learned.
• The output of that environment so far: nearly 800 shipped change requests.

The stack — what you will want to know

• TypeScript end to end: a NestJS API plus a separate NestJS worker on Railway; a Next.js 16 / React 19 frontend on Vercel.
• PostgreSQL with Prisma; Redis + BullMQ for queues and job orchestration; S3-compatible object storage; Pinecone for vector search.
• Agents are LangChain.js + LangGraph ReAct agents on Anthropic Claude models in 3 tiers, with OpenAI embeddings.
• Prompts are the single source of truth: every prompt body and model config lives in 1 versioned master file, synced to LangSmith — never hardcoded in TypeScript. The codebase stemmed from the prompts, not the other way around.
• Total capture: every model input, output, thinking block and tool call is persisted — observability, rollback evidence and future training data in one discipline.

Who I need

• A CTO partner for a product already in production. Two skills, in one person:
  • An experienced developer who has run software in production — deployments, incidents, the operational discipline of a product paying customers depend on.
  • A black-belt prompter. These are not the same skill, and we know it.
• The hint: if you watch developers work in AI environments and feel genuine frustration — everyone prompts, most prompt badly — and you have quietly built your own systems of rules, context discipline and reusable prompts, this posting is for you.
• You will run the product codebase and this environment, in Cursor, every working day.

The offer

• Generous stock options. Start-up austerity salary.
• A genuine partner's chair — empty since day one.

How to apply

• Apply here — 1 question at a time, the way our own product asks. Bring links: LinkedIn or CV, GitHub or portfolio.
• Or write directly — ahmet@boss-iq.com — and show me how you prompt. I read everything myself.

Apply — CTO Partner

Ahmet ÖzdemirFounder, BOSS-IQ