Hi Tip-Sheeters,

This week I'm bringing you a chat with another one of the speakers from the API Superstream I spoke at a few weeks ago. Zdenek Nemec is the founder and CTO of Superface.ai, where he is pushing the boundaries of agentic AI. He is a frequent speaker on APIs and AI, and presented at the Superstream on the topic “APIs for AI: Have we failed?”
​

Ryan: Your topic had an intriguing title “APIs for AI: Have we failed?” I think the first thing we all want to know is…have we?

Z: We (as API providers and product owners) are very slow at responding to the AI “movement”. We haven’t acknowledged the fact that there is now a new AI consumer and AI integrator on the block – the AI. And it has very different needs to the existing ones. Previously the integrators were humans and consumers of traditional deterministic software that had only few hardcoded ways to call APIs. The AI changed it all and it needs a very different approach from all from the user, user interface to the APIs. As a matter of fact it APIs are changed from “programmable” interfaces to interfaces for AI, where, eventually, nobody will be programming it.

This oversight with slow (if any!) reaction from the API community also contributed to the rise of MCP and screen controlling agents.

Bottom line: To be fair, there were some fast and good reactions like the AI SDK from a company dear to the API practitioners - Stripe. But those were rare.

Ryan: you are working on real-world implementations of APIs being consumed by Agents. What are you seeing as the top two or three tips you can give to API producers?

Z: Acknowledge the AI is the consumer of your API and treat it as a first-class citizen. Understand that in ten years time it might be the only consumer and integrator of your API. As with any good API the AI-ready API needs to understand its users, their business cases, needs and limitations. This is where a product owner should enter the stage. Providing an MCP server does not do you any good if the number of tools in it is over 20 and they are poorly documented.

Finally, the documentation is still the key, but there is a twist: it needs to be tailored for AI.

Ryan: how do you see this space developing in the next 6-12 months? What major advances or trends do you see on the horizon?

Z: We might see an explosion of MCP servers coming directly from the API providers. Take a look at the GitHub or Hubspot MCPs, for example. Which would marginalize the direct use of APIs by AI. The MCP, however, has many problems like authentication and authorization and needs to address them otherwise we will see another layer on top of the MCP.

Alternatively, the MCP “dream” won't materialize. Many companies (API providers) are just testing it to figure out if jumping on the MCP train will bring them more traction and/or stickiness. In this scenario we would need to double down on API design for AI as we discussed with Emmanuel Paraskis in our Maven Session (https://maven.com/p/d4116d/ai-agents-are-coming-for-your-api-are-you-ready)

Either way, AI will bring a very different type of consumption to your APIs and every provider needs to get ready for it.

Looking further, fully autonomous systems and agentic business will start to emerge as agents too will become API providers. This will need a whole new set of protocols which might start to emerge in the next 12 months.

Ryan: Speaking to individual data scientists and developers, what skills do you suggest they spend time developing this year?

Z: When you are an API provider you must have the understanding how LLM function calling works and how it relates to APIs. What are the limitations of LLMs and how your API (MCP) design affects the agentic success rates.

Right now, we are only scratching the surface with AI. Any production-grade agentic deployments will need to address the AI goal completion rates, reliability and governance. Understanding the AI systems and their limitations will give you the edge building AI-related products.

​

Good stuff from Z - to get more of his content, follow him on LinkedIn: https://www.linkedin.com/in/zdne/​
​

Keep coding,

​

Ryan Day

👉 https://tips.handsonapibook.com/ -- no spam, just a short email every week.

​

​