Antonio Ramos
AI agents, automations and data pipelines that ship fast.
New to Contra
Antonio is ready for their next project!
Tokens are cheap right now for the same reason Facebook ads were cheap in 2014. Somebody is buying market share.
That phase ends. It ended in paid media, it ended in cloud, it ended in streaming. Consolidation arrives, the subsidy stops, and the price finds its real level. Inference bottoms out at GPUs and electricity, not software margin, so its real level is not low.
The question I now ask about every automation I build: does this survive tokens costing 10x?
• n8n for anything that doesn't need judgment. Triggers, parsing, API calls, file moves, retries. It can't be repriced if it never calls a model.
• MCP for context. The model pulls what it needs through tools instead of swallowing a 40k dump on every call. Same answer, a fraction of the tokens.
• OpenRouter so no lab is load-bearing. Price triples, you change one string, you route to whoever stayed cheap. Lock-in is what makes a price hike stick.
Today that stack buys you margin. After consolidation it buys you a business.
Cheap tokens are a phase. Architecture is permanent.
2
105
Document extraction with a confidence gate: AI that escalates instead of guessing
0
17
AI agent that writes, scores and rewrites its own ad scripts
0
23
Competitor ad teardown: Apify to vision AI to scored report
0
28