Constructing Blocks for Agentic AI: The Hierarchy of Developer Wants

Because the Mannequin Context Protocol (MCP) was introduced by Anthropic a 12 months in the past, we’ve seen immense development in giant language fashions (LLMs) and agentic use circumstances. Earlier than MCP turned the de facto agentic normal, builders constructing brokers on high of LLMs must hard-code the connective tissue between the LLM and apps. Builders would wish to construct customized integrations between their LLM consumer and the apps required by an finish person’s immediate. With MCP, builders can now join immediately with exterior knowledge sources, so their LLM can learn knowledge from and write knowledge to the related functions.

However there’s a breaking level the place issues begin to collapse. The robustness and efficacy of agentic options depend upon the standard of the applying programming interfaces (APIs) which are utilized by an MCP server. MCPs expose instruments which are invoked by LLMs, and these instruments usually reference particular person API endpoints. The standard of APIs, subsequently, immediately correlates with the LLM consumer’s correct discovery and execution of person prompts.

The hierarchy of developer wants

Delivering high quality, constant, and intuitive APIs is a matter of a company’s cultural priorities. At Cisco, we’re lucky to have govt management sponsorship of our platform API floor space, which helps us ship high quality APIs throughout the enterprise.

We’ve discovered that it’s useful to prioritize API roadmaps based mostly on a hierarchy of developer wants.

Maslow’s hierarchy of wants illustrates the idea that human motivation is pushed by 5 ranges of want. Organized in a pyramid, these are: physiological, security, love/belonging, esteem, and self-actualization. The idea means that individuals are motivated to meet primary wants first earlier than progressing to higher-level wants for self-fulfillment.

Our hierarchy of developer wants takes Maslow’s pyramid and extends it to rank developer ache factors and options in keeping with primary wants, enterprise wants, and vertical or industry-specific wants.

The foundational parts: API protection and completeness

The developer hierarchy of wants provides a framework for our product and engineering groups to prioritize their roadmaps. For instance, the inspiration of API high quality relies on protection or completeness. With out extant API endpoints, nothing upstream issues—not documentation, code samples, nor developer relations efforts.

Determine 1: The hierarchy of developer wants provides a framework for product and engineering roadmap prioritization.

Subsequent comes the necessity for a high-quality OpenAPI Specification (OAS) and the inner engineering API growth lifecycle. An OAS file may have a lot of APIs, but when the prose round every endpoint is sparse, or sure endpoints are lacking schemas and documented parameters, then it forces builders to both guess and check or reverse engineer which capabilities every API endpoint serves by rote testing. Each choices provide a poor developer expertise.

It’s necessary for any group with a developer movement to spend money on linting and static evaluation of latest APIs being developed, in order that inner engineers can simply ship new API performance with out being consultants within the API framework. An excellent instance of that is Cisco DevNet’s open-source instrument, API Insights, which permits builders to audit their API specs to implement finest practices as early as attainable within the growth cycle. Additionally, many individuals aren’t conscious that if the OAS is lacking schemas, it basically means software program growth kits (SDKs) and infrastructure as code (IaC) instruments like Terraform and Ansible can not use these endpoints—no matter their launch standing—since SDKs and IaC instruments require the flexibility to auto-generate tooling from the OAS.

The ethical of the story: All issues depend upon OAS, particularly AI.

Documentation and scalability

The following stage of significance is having intuitive documentation. In case your APIs are improbable however nobody can discover which one does what, or your data structure is a nightmare, it negates the trouble your crew put into transport high quality APIs and OAS. Builders aren’t monoliths. Some are profession software program engineers, however many, particularly within the networking {industry}, are technical professionals like networking engineers who’re studying the way to write scripts or construct primary internet apps utilizing REST APIs for the primary time of their profession. Accordingly, nice developer documentation is a must have earlier than anybody can perceive what is feasible for an AI agent use case.

Transferring to the following stage of developer wants, we now have enterprise scale and high quality. Organizations that promote to Fortune 500 firms or brownfield enterprises know that scale and the CIA-triad (confidentiality, integrity, availability) are mission-critical tenets. These tenets should be assured earlier than builders will really feel snug constructing customized options on high of an API, no matter whether or not these options are conventional scripts, app integrations, or agentic troubleshooting options.

At Cisco, we’re proud to carefully companion with our largest clients and managed service supplier companions to make sure that our platform works easily for everybody, from a mom-and-pop brick-and-mortar store to a Fortune 100 international conglomerate that wants redundancy, immense scale, and price limits, in addition to the very best uptime out there.

Simplifying API consumption and vertical software

Our penultimate stage of developer wants is the place we get into abstraction and obfuscation to simplify how APIs are consumed. SDKs and IaC instruments like Ansible and Terraform are the bread and butter of enormous enterprises that require widespread automation. IT groups are rising leaner, and networking engineers are anticipated to do extra with much less. If all different developer wants have been met, it turns into straightforward for a company to generate and preserve SDKs and developer instruments like an MCP server.

For instance, Cisco Meraki and Cisco Catalyst Middle each preserve high quality OAS recordsdata, which clients can use to generate an MCP server inside a couple of minutes with common API instruments like Postman. This permits a developer who’s attempting to construct a customized agent to have the ability to achieve this in days, not weeks or months, all due to the standard of our APIs, OAS, and docs, and our means to make sure that high quality at enterprise scale.

Lastly, what actually makes a developer platform shine are code samples and integrations. If all different developer wants are met, this creates an ecosystem situation the place builders can simply create new scripts, integrations, and instruments and share them with the platform neighborhood within the type of code samples, libraries, and integrations.

Finally, delivering full, high-quality APIs, strong documentation, and an engaged ecosystem lays the groundwork for innovation and ensures that agentic options can attain their full potential.