What I Believe About AI, Software and Business

This is an opinion piece. It sets out the beliefs I bring to every product decision, consulting conversation and business idea. They are not universal truths; they are the conclusions I have reached from building and studying business software.

1. Most AI value is in unglamorous places

The demos that get attention are creative: image generation, chat, code. The value in most businesses is in reading invoices, triaging enquiries, summarising calls and cleaning data. These tasks are high-volume, measurable and dull — which is exactly why they are worth automating. If you run a business and want to start with AI, start with the work your team does on autopilot. I expand on this in AI Automation.

2. A model is not a product

Calling a model API is easy. Building something people rely on is not. The product is the workflow around the model: the context you give it, how you validate what it returns, how a person corrects it, what gets logged, what happens when it fails. Teams that skip this layer ship impressive prototypes and abandoned features.

3. Deterministic where certainty matters, AI where flexibility matters

Money movement, permissions, compliance checks and audit trails should be written in plain code. Reading a messy document, understanding intent or drafting text is where a model earns its place. The architecture question is always: where is the boundary?

4. Problem first, then technology

The worst software I have seen was built because a technology was exciting. The best was built because a specific person's specific problem was understood deeply. That is why my business ideas start with the problem and the customer, and only then describe the technology's role.

5. Validation beats conviction

Founders love their ideas — I love mine. The antidote is cheap tests: do the task manually for a few real users, charge a small amount early, watch what people actually do. I have written a full framework in How to Validate an AI Business Idea.

6. Vertical beats horizontal for small teams

A general-purpose AI tool competes with the largest companies in the world. A tool for accountants who serve small manufacturers in one country competes with almost nobody, and can go deep enough that AI becomes invisible. Depth in a workflow is a moat; a clever prompt is not.

7. Build to swap

Model capabilities and prices change every few months. Any AI product should be able to change providers without a rewrite. Practically: a thin model-calling layer, versioned prompts, and an evaluation set that tells you whether the new model is actually better for your task.

8. Honesty compounds

No invented case studies, no guaranteed outcomes, no keyword-stuffed pages. Say what is an idea and what is live. Label opinion as opinion. It is slower, and it is the only strategy that still works in five years.

9. India is a builder's market

Many processes in Indian businesses are still manual, mobile-first and multilingual. That is a design constraint and an enormous opportunity. I write about it in AI Business Opportunities in India.

10. Execution is the idea

Ideas are cheap and I publish mine freely. What matters is choosing one, scoping it down to an MVP, shipping it to real users and iterating on what you learn. Everything else on this site is in service of that.

If any of these beliefs resonate — or you think one is wrong — I would genuinely like to hear it. The contact page is open.