Eighty percent of Indian AI agencies have nothing in production. They pitch capabilities they've never shipped. Here's how to spot one — and what to look for instead.
Walk into any tech park in Bengaluru, Gurgaon, or Hyderabad. Half the doors say AI agency. Most of them have never shipped a production AI system. They're slideware shops — competent at decks, weak at code.
The four tells of a slideware shop
1. They lead with capabilities, not case studies
Their website lists everything: agents, vision, RAG, fine-tuning, voice, computer use, MLOps, MCP, A2A. No case studies. No screenshots. No mention of customers.
2. They quote in hours
Hourly billing is the leakiest contract in software. It transfers all risk to you and rewards slowness. Productized fixed-scope engagements transfer risk to the agency — where it belongs.
3. The proposal is sixty pages
Length signals defensiveness. The team that knows what they're building writes a one-pager.
4. No evals, no monitoring, no runbook
Ask: 'How do you know the agent is still working three months in?' If the answer is 'we'll write some tests', they don't run production AI.
What to look for instead
- —Live URL to a product they shipped
- —Three case studies with measurable outcomes
- —Eval methodology — they should explain it in plain English
- —Fixed scope, fixed price, sharp proposal in 48 hours
- —A handover deliverable — runbook, monitoring, retention plan
- —Their own product. Studios that build for themselves first know what they're doing
If they can't show you a system in production, they're selling you a story.