From Developer Experience to Agentic Experience: The Next Evolution of APIs

The emergence of AI agents is changing the way we consume and integrate APIs. We no longer design platforms just for developers, but also for agents capable of discovering documentation, using SDKs, running CLIs, and orchestrating workflows autonomously. In this talk, we’ll explore how a strong Developer Experience (DX) strategy forms the foundation of an excellent Agentic Experience, what this means for platform teams, and why organizations that invest in DX today will be better prepared to lead the era of agentic AI.

Transcript

1. From Developer Experience to Agentic Experience The Next Evolution of APIs Benjamin Granados benjamingm.xyz
2. benjamingm.xyz - 2
3. 75% of AI tasks fail when they hit an API Agents are already calling your APIs. Most fail. Not because the agent is broken, because the API was never designed to be understood by one. ZDENEK NEMEC - APIS AT THE VERGE OF AI (2025) benjamingm.xyz - 3
4. ABOUT ME Benjamin Granados Head of Global AI Community - Banco Santander MIT Faculty Collaborator - Digital Platforms Course Technical Evangelist with 25 years of experience leading initiatives across: KEY AREAS IN THE ECOSYSTEM • • • • • • • Speaker at: • PyConES • API Days • API Addicts Days • Codemotion • OpenExpo • T3chFest API Strategy API Governance API Products Adoption Open Source Open Standards Communities Developer Relations benjagranados benjamingm.xyz benjamingm.xyz - 4
5. Agenda 1 Developer Experience: The foundation. 2 The shift: APIs consumed by agents, not only humans. 3 Agentic Experience framework. 4 AX beyond APIs. 5 What is next? benjamingm.xyz - 5
7. FOUNDATION What is Developer Experience? DX = The overall quality of interactions of a Developer with a product, platform or system while achieving their jobs to be done. Docs Sandboxes SDKs & tools Examples, quickstarts, scannable. Docs as Code. TTFC minimum, sandbox ready. Native, typed, intuitive CLI. Consistency Community Trust Specs, Naming, versioning, patterns. Forums, examples, feedback. SLAs, errors, observability. benjamingm.xyz - 7
8. INDUSTRY DX in the industry: the API as a product DX turned the developer into the customer. The goal: reduce Time to First Call (TTFC). THE REFERENCES Stripe THE PRODUCT LOGIC Postman Cloudflare Canva Twilio GitHub • Developer Portals: Interactive docs & sandboxes as the primary UI. • Specs as Product: OpenAPI/AsyncAPI as living contracts, not just docs. • Consumption-First: Focus on immediate value realization. Slack These companies didn't just build APIs; they built a Developer Experience. They built these portals for humans. benjamingm.xyz - 8
9. SHIFT Agents are already calling your APIs Postman - State of API Report 2025: 5.700+ respondents. APIs now feed agents too. 24% 60% 10% 89% design APIs with AI agents in mind still design primarily for humans only use MCP regularly despite 70% awareness of developers use AI in daily work benjamingm.xyz - 9
10. DEFINITION What is Agentic Experience? AX = Encompasses how easily agents can access, understand, and operate within a product, platform or system to achieve user-defined goals. DX — THE DEVELOPER AX — THE AGENT • • • • • • • • Reads docs, infers intent. Interprets ambiguity. Keeps context. Engage support channels if finds blockers. Don’t read docs. Executes intent. Zero tolerance for ambiguity. Chains calls, loses context. Breaks silently. Then disappears. Same Developer Product. Two completely different consumers. benjamingm.xyz - 10
11. AGENTIC EXPERIENCE · THE OPERATIONAL LOOP The Agent Journey The operational lifecycle of Agent when evaluating Developer Products 01 DISCOVERY 02 CAPABILITY MAPPING Find the operation Match the intent The agent scans MCP servers or OpenAPI specs; it does not read a landing page. It analyzes descriptions and enums: does this API actually solve the user’s intent? 03 EXECUTION 04 Identify next step Build the request It builds the call from the schema constraints and invokes the API. ERROR RECOVERY It parses the error to decide whether to retry, pivot, or stop. 05 VALIDATION Verify the outcome It checks the payload against the expected goal before reporting back. The AX Friction Points DISCOVERY FAILURE MAPPING FAILURE RECOVERY FAILURE Signal: poorly named endpoints or missing descriptions. Signal: ambiguous parameters lead to “hallucinated” inputs. Signal: generic 400 errors leave the agent stuck in a loop. KEY TAKEAWAY · “AX is not a single feature; it is the optimization of this entire journey to ensure the agent reaches the ‘Validation’ step without human intervention.” benjamingm.xyz - 11
12. AGENT TYPES Two kinds of agent: coding vs autonomous They look alike from the outside, but what they need from your API is different. Coding agents Autonomous agents • • • • Goal: write, refactor, integrate or explain code that calls your APIs. Needs: clear schemas, typed examples, SDK signatures, error patterns. Failure mode: hallucinates parameters because the spec is ambiguous. • • Goal: execute tasks end-to-end with real side effects. Needs: intents, workflows, guardrails, non-human identity, observability. Failure mode: calls the right endpoints in the wrong order, or with the wrong context. benjamingm.xyz - 12
13. FRAMEWORK The five shifts AX demands An agent does not browse your portal. It must discover what you offer, know the safe sequence to call, authenticate as a non-human, trace every decision, and stay inside your governance. 01 02 03 04 05 Discovery Workflows Security Observability Governance What can this API/Product do? What is the deterministic call sequence? Who is this agent and what can it do? What did the agent decide and why? What this agent is allowed to do? Postman 2025: 24% design APIs with agents in mind. benjamingm.xyz - 13
14. ASSESSMENT How do you measure AI-readiness? Jentic's AI Readiness Scorecard scores any OpenAPI document across six dimensions and returns a single grade with a prioritized fix list. • • • Foundational Compliance Developer Experience AI-Readiness • • • Agent Usability Security AI Discoverability Web: jentic.com/scorecard CLI: npx @jentic/api-scorecard-cli Github:jentic/jentic-api-scorecard Github Action & Agent skill benjamingm.xyz - 14
15. IMPLEMENTATION Beyond traditional APIs Agents need more: intent, workflows, skills, and runtime context. LLMS.TXT AGENTS.MD MCP / SKILLS A concise, machine-readable index of what your API or product does. Tells the agent where to start reading. A simple, open format for guiding coding agents. Tell the agent how to work on your project. The README.MD for agents. Expose capabilities as tools an agent can discover and invoke. A skill is a bounded capability, not a raw endpoint. ARAZZO WORKFLOWS Multi-step, deterministic business flows. Not what each endpoint does — how to sequence them safely. The goal: move from "here are my endpoints" to "here is what an agent can achieve." benjamingm.xyz - 15
16. IMPLEMENTATION LLMS.TXT # Twilio Documentation > Reference documentation for all Twilio products. Code snippets, tutorials, and sample apps for common use cases and communications solutions. ## Building with AI * [Building with AI](/docs/ai.md): Integrate Twilio APIs with AI coding agents using MCP servers, or use AI-powered communication products. Build faster with intelligent tools. * [Twilio MCP server](/docs/ai/mcp.md): Connect AI coding agents to Twilio's API documentation with the Model Context Protocol server. Search over 1,800 endpoints and retrieve schemas on demand. * [Twilio Skills for AI coding agents](/docs/ai/skills.md): Install Twilio Skills to teach AI coding agents which Twilio products to use, proven architecture patterns, and how to avoid common mistakes. Works with Claude Code, Cursor, Codex, and other Agent Skills-compatible tools. benjamingm.xyz - 16
17. IMPLEMENTATION Beyond traditional APIs Agents need more: intent, workflows, skills, and runtime context. LLMS.TXT AGENTS.MD MCP / SKILLS A concise, machine-readable index of what your API or product does. Tells the agent where to start reading. A simple, open format for guiding coding agents. Tell the agent how to work on your project. The README.MD for agents. Expose capabilities as tools an agent can discover and invoke. A skill is a bounded capability, not a raw endpoint. ARAZZO WORKFLOWS Multi-step, deterministic business flows. Not what each endpoint does — how to sequence them safely. The goal: move from "here are my endpoints" to "here is what an agent can achieve." benjamingm.xyz - 17
18. IMPLEMENTATION AGENTS.MD AGENTS.md — working in this repo with an AI coding agent This is the cross-tool instruction file. OpenAI Codex, Google Antigravity, Cursor, and other agents read AGENTS.md automatically. Claude Code users get the same guidance from CLAUDE.md / the skills/ directory. Keep this file tool-agnostic. What this repo is A complete, multi-cloud Salesforce Agentforce reference org for one (fictional) coffee business — Commerce (D2C + B2B), Service, Field Service, Revenue Lifecycle Management (RLM), Data Cloud, and Payments — with 12 AI agents on top, deployable to a fresh Developer/scratch org. See README.md for the full breakdown. The golden rule: DX-first, browser last When making changes, always prefer in this order: Salesforce DX / sf CLI — sf project deploy/retrieve, sf data, sf apex run, sf agent commands. Salesforce MCP tools (if connected) — deploy_metadata, run_soql_query, run_apex_test, benjamingm.xyz - 18
19. IMPLEMENTATION Beyond traditional APIs Agents need more: intent, workflows, skills, and runtime context. LLMS.TXT AGENTS.MD MCP / SKILLS A concise, machine-readable index of what your API or product does. Tells the agent where to start reading. A simple, open format for guiding coding agents. Tell the agent how to work on your project. The README.MD for agents. Expose capabilities as tools an agent can discover and invoke. A skill is a bounded capability, not a raw endpoint. ARAZZO WORKFLOWS Multi-step, deterministic business flows. Not what each endpoint does — how to sequence them safely. The goal: move from "here are my endpoints" to "here is what an agent can achieve." benjamingm.xyz - 19
20. IMPLEMENTATION SKILLS VS MCP benjamingm.xyz - 20
21. IMPLEMENTATION Beyond traditional APIs Agents need more: intent, workflows, skills, and runtime context. LLMS.TXT AGENTS.MD MCP / SKILLS A concise, machine-readable index of what your API or product does. Tells the agent where to start reading. A simple, open format for guiding coding agents. Tell the agent how to work on your project. The README.MD for agents. Expose capabilities as tools an agent can discover and invoke. A skill is a bounded capability, not a raw endpoint. ARAZZO WORKFLOWS Multi-step, deterministic business flows. Not what each endpoint does — how to sequence them safely. The goal: move from "here are my endpoints" to "here is what an agent can achieve." benjamingm.xyz - 21
22. IMPLEMENTATION ARAZZO SPEC arazzo: 1.0.0 info: title: Pet Store Adoption Workflow version: 1.0.0 description: A workflow to search for a pet and create an adoption order. sourceDescriptions: - name: petstore-api url: https://swagger.io type: openapi workflows: - workflowId: adoptPetWorkflow summary: Search for an available pet and place an order inputs: type: object openapis.org/arazzo-specification benjamingm.xyz - 22
23. EXAMPLES Two agents, two needs CODING AGENT AUTONOMOUS AGENT AUTONOMOUS AGENT Twilio - Forward thinking Robinhood - MCP based Support both Agent types: Coding and Acting. Advanced top documentation and resources. Executes a trade end-to-end with real money at stake using MCP. benjamingm.xyz - 23
24. THE EVOLUTION CONTINUES The headless platforms salesforce.com/news/stories/salesforce-headless-360-announcement benjamingm.xyz - 24
25. THE EVOLUTION CONTINUES AX beyond APIs latecheckout.agency/ax-book benjamingm.xyz - 25
26. NEXT STEPS Start today! Two concrete things to try out: 1 Model one workflow in Arazzo and pass it to an AI Agent. 2 Score one API with Jentic's AI Readiness Scorecard. usearazzo.com/ecosystem/#tools · jentic.com/scorecard benjamingm.xyz - 26
27. From Developer Experience to Agentic Experience The Next Evolution of APIs Benjamin Granados benjamingm.xyz Thanks!
28. RESOURCES Sources & further reading POSTMAN STATE OF THE API 2025 FRANK KILCOMMINS - PLATFORM SUMMIT 2025 postman.com/state-of-api · 5,700+ respondents. Beyond the Agent Hype JENTIC AI READINESS SCORECARD OPENAPI - Arazzo Specification jentic.com/scorecard · 6 dimensions, CLI, skill, action. openapis.org/arazzo-specification ZDENEK NEMEC - API the Docs 2025 SALESFORCE HEADLESS 360 Apis at the verge of AI salesforce-headless-360-announcement benjamingm.xyz - 28