Most people pick one AI model and use it for everything. ChatGPT for writing. ChatGPT for research. ChatGPT for code review. ChatGPT for data cleanup.
That's like hiring a senior engineer to do data entry. It works — but you're overpaying for every task that doesn't need that level of firepower.
The Workflow Lab: Model Routing for Non-Engineers
Here's the concept: different AI tasks have different complexity levels. A quick summary doesn't need the same horsepower as a 2,000-word article. A trend scan doesn't need the same model as a deep competitive analysis.
We run 5 models across 3 tiers:
Tier 1 — Free (local models via Ollama)
These run on a Mac Mini sitting in the office. No API costs. No monthly subscription. Tasks like trend scanning, quality scoring, data parsing, and code debugging all run here. Models: qwen2.5, deepseek-r1, llama3.1.
Tier 2 — Cheap API calls
For quick drafts, summaries, and format conversions. Claude Haiku, GPT-4o-mini — these cost fractions of a cent per request.
Tier 3 — Premium API calls
Production content, complex reasoning, client deliverables. Claude Sonnet or GPT-4o. This is where quality matters most, so this is where you spend.
The key: only about 15-20% of tasks actually need Tier 3. The rest run free or nearly free.
How This Actually Works in Practice
Our content pipeline is a real example. Four AI agents run in sequence through n8n (a free, open-source automation tool):
1. Scout scans for trending topics — runs on a free local model
2. Analyst verifies claims and adds depth — runs on a free local model
3. Creator writes production content — runs on Claude Sonnet (paid)
4. Director scores quality — runs on a free local model
One paid step. Three free steps. Total pipeline cost: $0.21. Output: a newsletter issue, LinkedIn post, X thread, and video script.
The Takeaway
You don't need to switch AI providers. You don't need to cancel everything. You just need to stop routing simple tasks through expensive models.
If you want to try this yourself, start with Ollama (ollama.com). It's free, installs in 5 minutes, and runs models locally on your machine. Route your testing, debugging, and drafting through it. Keep your paid subscription for the work that actually needs it.
MEWR Creative runs 22 AI agents across 8 departments. We write about what we build, what works, and what doesn't. If you got value from this, forward it to someone who's overpaying for AI.