Agentic AI, without the hype
What it actually takes to put an AI agent inside a website or product — and where we think it genuinely helps.
Every brief we’ve read this year mentions AI. Most of the time, what the client actually wants isn’t a chatbot in the corner of the screen — it’s for their product to do more of the work for their customers. That’s the promise of agentic AI: software that doesn’t just answer, but takes steps on your behalf.
We’re a small studio, and we’re still learning this alongside everyone else. These are the principles we’ve settled on so far.
Start with a job, not a model
The useful question isn’t “where can we add AI?” but “what tedious, multi-step task are people doing here?” On an e-commerce site that might be narrowing two hundred bottles down to the three that suit a gift and a budget. In a booking flow it might be comparing dates, prices and policies. If you can’t name the job in one sentence, an agent will only add noise.
Agents need a good interface too
An agent is only as good as the actions it can take. That means clean, structured product data, clear APIs for search, cart and checkout, and permissions that are deliberately narrow. Most of the work in a good AI feature is ordinary engineering: tidy data, well-defined tools and sensible limits.
Show the work, keep people in charge
People trust an agent when they can see what it’s doing. We design the steps to be visible — “searching single malts under ₹8,000… comparing three…” — and we keep a human in the loop for anything that costs money or can’t be undone. A confirm button is a feature, not friction.
- Explain what the agent is about to do before it does it.
- Let people edit, undo or take over at any step.
- Say clearly when it isn’t sure — and hand over to a person.
Design for agents as visitors, too
There’s a quieter shift happening: more of your visitors will be AI assistants browsing on someone’s behalf. Semantic HTML, honest structured data and fast, accessible pages help those agents understand you — exactly as they help people. Good craft turns out to be good AI-readiness.
The best AI features don’t feel like AI. They feel like the product finally understood what you were trying to do.
If you’re thinking about an agentic feature and want a practical second opinion, we’d love to hear about it.