Issue #87

Top engineers at Anthropic and OpenAI say AI now writes 100% of their code.

If you're not using AI, you're spending 40 hours doing what they do in 4.

These 100+ Claude Code hacks fix that and help you ship 10x faster.

Sign up for The Code and get:

Claim your free playbook

As agent skills become a way of packaging domain knowledge for AI agents, the challenge quickly shifts from creating them to keeping them accurate, consistent and reliable at scale. Google Cloud shares how it approaches this problem across its growing Agent Skills library.

AI agents can access tools, APIs and services through MCP, but access alone does not teach them how your organisation expects those capabilities to be used. This practical guide explores Agent Skills as a way to package procedural knowledge into portable, version-controlled instructions that agents can discover and apply when needed.

Developer portals were built for humans: browse the documentation, test an endpoint, generate a key and start building. But AI agents increasingly discover and consume APIs in very different ways. Kristopher Sandoval explores what an AI-native developer portal might look like, from machine-readable capabilities and workflow context to runtime discovery and better MCP integration. It also tackles an important question: if MCP can expose APIs to agents, does the developer portal still matter?

As APIs evolve from individual endpoints into multi-step workflows, governance needs to cover more than the underlying API definitions. ​​Frank Kilcommins explores how Spectral can bring linting and style-guide enforcement to Arazzo workflows, helping teams validate consistency and enforce their own rules. The bigger opportunity is making API workflows machine-readable, predictable and governable, particularly as AI agents become consumers of those workflows.

As generative AI accelerates development, API changes can outpace the tests and documentation designed to keep them in check. Ewa Ostrowska put its own drift-detection capability to the test on a large, mature API, and uncovered mismatches between the OpenAPI definition and the behaviour of the running service. What makes the case study interesting is that the checks exposed more than documentation gaps: they also uncovered genuine bugs and authentication ambiguities that existing tests had missed.

As organisations move from experimenting with AI to deploying agents and AI-powered applications at scale, governance becomes increasingly difficult to manage. Brendan Bondurant & Tanya Deputatova explore the frameworks organisations can use to bring structure to AI risk, security, compliance and accountability.

API authentication looks standard on paper, but operating integrations at scale reveals just how differently providers implement it. After supporting more than 900 APIs, Nango shares the production failures and edge cases that turned seemingly ordinary OAuth problems into serious reliability issues.

One of the biggest attractions of MCP is making existing APIs accessible to AI agents, but doing so traditionally means building and maintaining another server. Google Cloud is taking a different approach by allowing Google API Gateway to expose existing REST operations directly as MCP tools.

As enterprises rush to deploy AI agents, the biggest obstacle may not be the models themselves, but connecting them reliably to decades of existing systems and APIs. Jentic CEO Sean Blanchfield describes these accumulated layers of legacy technology as the “geological layers” of enterprise IT, and explains why they are becoming a major barrier to AI adoption.

Many production APIs still lack accurate OpenAPI descriptions, making them harder for developers, and increasingly, AI agents, to understand and use. Adam Sobaniec explores a different approach: reconstructing an API specification from actual HTTP traffic rather than trying to infer it from a large codebase.

When Stainless announced it was joining Anthropic and winding down its hosted SDK generator, existing SDKs did not suddenly stop working. The bigger problem is what happens the next time an API changes. Scalar’s account of the wind-down explores the often-overlooked dependency behind SDK generation.

What if exposing an existing API to AI agents did not require building another server? Google Cloud is now allowing API Gateway to act as a remote MCP server, turning existing REST operations into agent-ready tools using annotations in the OpenAPI specification.

The API Platforms For Scale Newsletter curates intelligence on the trends, technologies, and practices shaping API platforms.