APIDays 2023 Recap

To finish this year after having some nights of thoughts - I also wanted to summarize my impressions of this years great APIDays conference.

API Specifications

The OAI track at APIdays Paris led by Erik Wilde proved to be an incredible insightful experience, marked by the notable announcement of the brand new AsyncAPI spec v3 to ease the approach how we describe event-driven interfaces by decreasing the potential for misunderstandings and ambiguities but also enabling better reuse with new channel definitions between different specifications. Definite a must check topic. Another highlight was the outlook towards OpenAPI v4 (aka Moonwalk). Seems like the change of structures will allow us to further automate and support API spec handling with more tooling for even more complex use cases. Through the inclusion of JSON Schema approaches towards APIs, Event-driven with AsyncAPI but also parts of standards for data contracts – we might get better reusability and up-to-date specs through the end-to-end chain of dev teams. The more this goes hand in hand with the daily usage in a developer’s lifecycle, the better documented and maintained API specs hopefully will be.

The aggregated API example definitively looks like a common use case that we have when combining logical groupable microservices into a single use API fa?ade: https://github.com/OAI/moonwalk/blob/main/examples/aggregatedapi.yaml. But also https://github.com/OAI/moonwalk/blob/main/examples/rpc.yaml is maybe a bridge that could allow us to use other approaches other than classic REST APIs with the power of OAS tools. Looking through the GraphQL conference track on day 3 showed that there is maybe much more challenges and similarities independent of the chosen interface architecture or message protocols.

But coming back to the API specs topics - “Get your OpenAPI Spec Ready for Code Generation” by Nolan Di Mare Sullivan was immensely valuable by providing hands-on approaches and numerous practical insights on code and SDK generation questions. But even more interesting was the work on Speakeasy suggest: https://www.speakeasyapi.dev/post/release-speakeasy-suggest. That seems like a great support tooling for our API development teams.

Lastly the talk on managing OAS at scale by Stève Sfartz was a great summary on best practices and lifecycle recommendations around your API specification and design processes.

And finally good to hear that our community is again thinking about scaling the topic of API Discovery or at least bringing it back into scope. Really interesting talk by Steven Willmott on apis.json and how to build an API search engine. I believe merging this approach with some more automated tooling, together with leveraging on RFC 8615 defined by IETF, we could express much better Metadata on existing API endpoints.

Leveraging those docs additionally to standards like API workflows maybe really enables us to have autonomous API interactions using AI agents. OpenAPIChain in langchain is already quite impressive but still constraint to be prepopulated with specs and secrets.

?API the Docs

?This leads to another interesting track on API documentation and user-centered design. Good to see the talks like Polina Zaichkina s talk on “transforming legacy documentation” that used toolings and a CMS should not constraint how we serve docs to our users. And I absolutely love the attitude, that if your tools don’t match then start your own tool development teams, to provide better user experience and have more encouraged teams.

I hope I could convince some people that using modern Markups and docs-as-code is incredible useful and a Backstage.io portal definitely is ready for production (just saw another talk questioning about this somewhere else).

Another highlight again seeing Kristof Van Tomme ’s panel discussions on the future of dev portals and how they support Enterprise API programs. I totally share the impression that those will be even more crucial with a rapidly increasing amount of interfaces, be it APIs, Events or LLM-extensions. Many people in this area think about the AI bots leveraging human docs for accessing interfaces but I believe in the future it is also about developer docs for chat interfaces. We use so many copilot-likes and try chat and instruction LLMs and learn that proper prompt engineering is a science in its own. So I am just waiting for the swagger-ui for chat prompts.

Use case talks

?For me there was much value in attending or watching industry use case talks and experience sharing, for example the very interesting talk on "API Governance at L'Oreal" by Marina Passoubady and Liliana Jena . Especially as business and IT alignment in a not-just tech environment (like ours) is crucial to adopt APIs as products but also API and Data team alignments. So valuable to share some same experiences.

Same story with additional best practices how to close gaps between business and IT orgs in "Driving an API Mindset in a Complex Organization" from Paul Williamson . Really insightful story about a quite long maybe everlasting journey.

And of course, impressible insights into "Best-In-Class Logistics APIs" by Karsten Wagner . I am pretty sure if there is a negative equivalent of API latency in the physical world then the people at DHL are used to manage this.

Even if it was in the OAS track, for me the talk on “How OpenAPI Supports Airline Industry Standards and Beyond” by Yan Sun is another great example for great value using open and standardized APIs in an (so our) industry. Really keen to watch how this journey continues.

GenAI and LLMs

Felt this was the overriding theme of this year’s conference, you could feel the interest and a lot of use cases to improve API developer efficiencies, but I felt I still missed the very stunning examples, maybe after watching the Google Gemininis blue duck before the APIdays GenAI talks were just too introductionary. Don’t get me wrong I know that AI and GenAI actually arrived in our industry and will stay and reshape it for the next months.

?Two of the most outstanding examples on API development powered by LLM agents was the PolyAPI demo by Darko Vukovic and the already mentioned Speakeasy LLM approach. As said I still expect copiloting moving into majority of the developer toolsets. Another great presented approach was on “Cloud APIs, ChatGPT 4-Turbo, and Attack Path Visualization” by Doug Dooley – leveraging GenAI both on the good and bad site of light in future.

We saw talks using VectorDB services for creating LLM embeddings, some early nice demos and people talking about machine versus human-language based interfaces. Some talks in the booths about deterministic and non-deterministic access of interfaces (feels we are not beyond APIs but maybe the way of consumption and the resilience towards usage errors at high scale needs to improve) but not a common consensus. For me it still feels that the battle is not yet over, so OpenAIs API will not be the industry standard yet, but there is also no standard way to interact with any running and open language model instance, neither how to extend them across platforms. Maybe this takes more time but portability in the way we use generative AI interfaces, services, infrastructure and toolsets definitely feels still somehow fuzzy.

Learning and Training and API Security

Great to see more initiatives around API training and learning facilitation, of course there is tons of great youtube, linkedin and blog contents but new redacted learning paths and course selections seem very promising to help people with different skill sets master best practices and tools for API creation and development.

Outstanding of course again the talks from APISec University conference track (https://www.apisecuniversity.com), starting with Dan Barahona introducing the state of API security, of course Corey J. Ball when it comes about Hacking APIs and a great talk on API Authentication by Travis Spencer . Also mind-blowing was the live session on Hacking a Bank API by Edward Lichtner . So much great content to digest.

Tooling

The new API landscape now needs an index of its own laugh, but great to see it printed even and how much energy still is in this market: https://apilandscape.apiscene.io/

Green IO Conference

Finally a great overview on State of Tech Sustainability by Ga?l DUEZ . Good to hear that the industry is thinking a lot on carbon footprints and sustainability and maybe takes some of the measures and experiences we might have on API monetization topics to reuse for green io measures. Feels like we really need to take actions into account as with the great power of GenAI utilization comes heavy responsibilities to not just waste energy for fun bits. Yes we are all in on customer obsession, developer experiences and great user flows but also tech has its footprint.