Why Most AI Agents Lose Money and How Are You Pricing Expensive Agent Workflows

Hi PH community 👋

We’d love to get advice from AI Agent builders and practitioners who are deploying real AI agents.

We run a platform for AI agent Marketplace and deployment middleware and are shipping multiple agents ourselves. What we’ve discovered is concerning:

Many AI agent projects are quietly losing money.

The reason? High tool API usage (especially expensive image / rendering generation), Heavy LLM API calls, Multi-step workflows. Agents have real variable cost per run.

For example, we just launched the Craftsman Agent product

https://www.producthunt.com/products/craftsman-agent?launch=craftsman-agent

🎯 Our Case: Craftsman Agent

The agent turns any ideas(prompt or images) to 3D build instructions, such as prompt → LEGO / Minecraft-style assembly instructions → Step-by-step images → 3D render → (optional video). The agent workflow requires multiple times calling of expensive APIs.

💰 Real Cost Breakdown Per Each Workflow

Per full workflow run:

1. Assembly Step Images Generation: 1–10 images calling Gemini Nano Banana 2, ~$0.05–$0.10 per image, 5 step images on average, total ~$0.50

2. 3D Rendering API Rendering 4 angles: ~$0.50 per each run

3. Optional Video Generation (video of MOC assembly)

Total workflow cost per each run

👉 ~$1–$3 per run

This is real marginal cost. No “near-zero SaaS scaling.”

Pricing Strategy

In terms of pricing, we think a lot about the pricing strategy so not to lose money.

1. Free quota

How many free trials (1, 2-5?, more?) can each registered user have? So that we avoid keep losing money?

2. Option A - Pay Per Run/Pay Using Credit

Will $1.5-$4 dollars charge acceptable compared to the cost ($1 – $3)?

3. Option B - Subscription with Hard Cap

Free, Pro, Ultimate, like Pro plan $20 for 20 runs (cheaper than average per run)?, Ultimate $60 for $80 runs (we will keep losing money though...)?

Would love to hear from:

AI founders,Infra builders. Anyone who has struggled with variable inference cost

Anyone who figured out a sustainable pricing model?

Because right now, it feels like many AI agents are growing revenue… but not profit.

Looking forward to learning from the community 🙏

This is part of the 3D image that we are rendering

Step-By-Step Instructions