By Token costs are one of the least visible but fastest-growing line items in AI agent infrastructure. As context windows fill with tool call responses and raw API data, agents become more expensive to run and measurably less capable at reasoning. Teams that optimize data routing early see dramatic reductions in both spend and agent performance degradation.
In this interview on TFiR, Jean Lafleur, Co-founder and Chief Operating Officer at Airbyte, breaks down the token cost benchmarks Airbyte customers are seeing with smarter data routing compared to raw APIs and MCP Gateway.
Guest: Jean Lafleur, Co-founder and Chief Operating Officer at Airbyte
Show: The Agentic Enterprise
LinkedIn: Jean Lafleur
Watch the discussion on YouTube
Here is what every AI platform engineer and agent developer needs to know.
Technical Deep Dive
Q: How much do AI agent token costs drop with optimized data routing versus raw APIs?
Jean Lafleur, Co-founder and Chief Operating Officer at Airbyte, states that customers using Airbyte’s approach see context token usage drop to 10% of what raw API calls consume, a 10x reduction. This is significant not only for cost but also for agent reasoning quality, since a less saturated context window keeps the agent sharper throughout a session.
“Compared to raw APIs, we had 10x [fewer] context tokens used.”
Jean Lafleur, Co-founder and Chief Operating Officer, Airbyte
Q: How does MCP Gateway compare to raw APIs for AI agent token efficiency?
Lafleur reports that MCP Gateway performs better than raw APIs but still falls short of Airbyte’s routing approach. Customers see 40% fewer tool calls and half the context tokens compared to MCP Gateway, meaning MCP Gateway is a partial improvement, not a complete solution to the token cost problem.
“For the MCP gateway, we see 40% less tool calls in terms of token usage compared to raw APIs, we had 10x [fewer] context tokens used, and compared to MCP Gateway, we had twice less context tokens used.”
Jean Lafleur, Co-founder and Chief Operating Officer, Airbyte
Q: Why does context window saturation make AI agents less accurate?
As an agent’s context window fills with tool responses and data payloads, its reasoning degrades because the model has less effective capacity to process new instructions and maintain coherent chains of thought. Lafleur flags this as a compounding problem: cost rises while agent intelligence falls simultaneously, making context efficiency a dual performance and cost issue.
“Your agent becomes less smart while the context window is being filled.”
Jean Lafleur, Co-founder and Chief Operating Officer, Airbyte
Resources & Documentation
- Airbyte, open-source data movement platform for AI agent data routing and pipeline optimization
***
👇 Click to Read Full Raw Transcript
Swapnil Bhartiya: One of the problem is also challenge is also cost. Can you talk about if there is any impact on cost that your customers are seeing with these new capabilities?
Jean Lafleur: Yes, because they use a lot less tokens. That’s actually pretty significant. Depending whether they use we can compare to raw API calls or MCP Gateway. But for instance, for the MCP gateway, we see 40% less tool groups. Like fewer tool groups in terms of token usage compared to raw APIs, we had 10x. So with 10% in terms of context tokens used compared to raw APIs, and compared to MCP Gateway, we had twice less context tokens used that point. And so those are pretty significant in terms of cost, but also in terms of how in terms of context routing, where your agent becomes less smart while the context window is being filled.
