How I Ship AI Products Without Engineers

I spent thirteen years scaling growth for Amazon, L'Oréal and Nike — managing the people who built things. Today I build the things myself: three AI products, shipped solo, one already in paid beta. Here's the actual playbook, minus the hype.

The conventional wisdom says you need a technical co-founder to build software. For a long stretch that was true. It isn't anymore — and the gap between "had an idea" and "shipped a working product" has collapsed from months to days. What changed wasn't me suddenly learning to code like a senior engineer. What changed is that the tools now meet you where your judgment already is.

If you've run a P&L, scoped a roadmap, or argued a strategy deck to a sceptical board, you already have the scarcest skill in product building: knowing what is worth building and why. The rest is increasingly a tooling problem. Here's how I close it.

Start with the judgment, not the code

Every product I've shipped started as a one-paragraph problem statement, not a tech stack. Kinetiq began because I was tired of fitness apps that forgot my torn rotator cuff every single session. Headhunter began because job boards flood you with noise instead of the handful of roles actually worth your time. The strategy work — who is this for, what is the one job it does, where does it have to be excellent — is identical to what I did for brands. That part doesn't get outsourced to a model.

The scarce skill was never typing code. It's deciding what deserves to exist and holding the bar on quality.

Use AI as your engineering team, deliberately

I build with Claude Code as the primary engineer and myself as the product lead, reviewer and QA. The mental model that made this click: treat the AI like a fast, capable junior team that needs crisp direction and tight feedback loops. Three habits matter more than any framework:

• Specify like a PM, not a prompt-jockey. The quality of what you get back is mostly the quality of what you asked for. I write the acceptance criteria first.
• Work in small, verifiable increments. Ship one feature, run it, look at it with my own eyes, then move. Big-bang generations are where projects rot.
• Keep a human in the loop on anything that touches money, data or trust. The AI writes the code; I own the consequences.

The architecture I actually reach for

Under the hood these are multi-agent systems and RAG pipelines — not because they're fashionable, but because real problems need memory and specialised steps. Kinetiq has to remember your constraints across sessions; that's retrieval. Headhunter scores and ranks roles, then drafts tailored applications; that's a pipeline of focused agents, each doing one thing well. You don't need to invent this from scratch. You need to understand it well enough to direct it and to catch it when it's wrong.

What thirteen years of growth taught me that transfers

The non-obvious advantage of coming from strategy and growth isn't technical — it's everything around the code:

• Distribution is the product. I've watched brilliant features die for lack of a wedge. I scope every build with the go-to-market in the same breath.
• Unit economics keep you honest. Kinetiq is priced and costed before a line ships. Owning a ₹280Cr book teaches you to respect the maths.
• Taste compounds. Knowing when something is "good enough to ship" versus "embarrassing" is judgement, and it's the thing models still can't do for you.

If you're a non-engineer who wants to start

1. Pick a problem you personally feel every week. Motivation is the real bottleneck.
2. Write the one-paragraph spec and the success test before touching any tool.
3. Build the smallest version that does the one job, end to end. Use it yourself.
4. Show ten real people. Their reactions, not the model's confidence, tell you what's true.
5. Only then add the second feature.

None of this requires permission or a co-founder. It requires the willingness to ship something rough, look at it honestly, and improve it — which, if you've operated at any scale, is muscle you already have.