Spotify’s best engineers haven’t written any code since December — they don’t need to because AI agents are doing all the work. A stunning 75% of Google’s code is now being written by AI. And Booking.com measured a 16% boost in developer productivity after adopting AI coding tools.

AI agents are taking a growing number of tasks from developers’ plates. They can do a significant amount of engineering work without sleeping or needing breaks, and they’re pretty good at it. They’re refactoring codebases, writing test scripts, and chewing through back-end development like an army of college interns on Red Bull.

And that’s creating a different reality for software companies that work closely with developers as the primary users of their products. Now, AI agents are the ones visiting developer portals, going through onboarding, and combing through documentation. In short, agents are the new developers — and the portals and resources that were originally created for human consumption need to adapt quickly in response.

The API provider world is used to humans learning, onboarding, and integrating, as well as testing, deploying, and maintaining those integrations, said Alex Barnett, vice president of developer ecosystem and experience at Vonage. But now, companies need to make their developer platforms entirely available for agentic interaction so agents can perform those tasks. “This is a new wave of innovation for us,” Barnett added. “We’re restructuring our platform and services for the next era of agentic work.”

In many cases, AI agents are given assignments by developers and work autonomously to complete them. They reason through the assignments on their own, break them into smaller tasks, write and test code, and fix any errors along the way.

Developers are spending more of their time working within AI harnesses and agentic IDEs (Integrated Development Environments), which are built on top of large language models (LLMs) and add tools, context, and safety guidelines. Claude Code is a popular harness, for example, that can directly read codebases, edit multiple files, execute shell commands, and interact with the GitHub platform, allowing it to automatically execute large development tasks from start to finish.

These powerful AI-powered tools are transforming software development from manual programming to high-level systems thinking and agent orchestration. Instead of writing code line by line, developers can now focus on steering, prompting, and verification — acting as architects while the AI handles the bulk of tactical implementation. Those agents will assemble the context needed and pull internal files, developer instructions, skill definitions, and tool descriptions. Basically, the agents look for everything they need to start coding, and they do this within minutes.

“It’s hard to overestimate the shift that’s happening in software development,” wrote two senior partners at McKinsey who recently studied how AI is shaping enterprise software development. At the top-performing companies, AI adoption has led to improvements of up to 30% in productivity and 45% in software quality. “This change is leading to smaller teams, much lower unit costs for software development, and much faster idea-to-impact cycle times,” they added.

You can see why the amount of agent-generated traffic on the internet is exploding. A new report from Cisco described AI agents as “network power users,” and found that an AI agent on a task will generate 450% more total traffic than a human performing the same task. “Approximately 70% of that traffic is AI inference, making inference paths mission-critical,” Cisco researchers noted.

Enterprise traffic was expected to grow by 2.5x over the next decade without agentic AI, the Cisco report said. But when you add agentic AI to the mix, traffic could grow by 9x, fueled by autonomous task execution and inference-heavy workflows.

Developer platforms are now trying to understand how to give AI agents the right information, while also exposing the end points needed to enable agents to interact with API platforms directly. Even something that seems fairly insignificant, such as a tool description, will be analyzed instantaneously by AI to determine usefulness. That means every description or skill definition needs to be reviewed.

Barnett at Vonage gave two examples of specific changes. An API now should be designed, implemented, and documented so that an agent can start developing code against it. Secondly, the traditional developer dashboard must evolve to provide more functionality than humans need.

“The upshot is everything that can be done via a platform should be exposed via APIs, so the agent can interact with the platform completely on behalf of the human developer,” Barnett added. “For us as a developer platform, that’s an urgent transformation that we are undertaking ourselves.”

Nothing like a little pressure. So what can companies do to prepare their systems and documentation for agentic developers?

“Think about the agent experience,” Barnett said. “We need to be as thoughtful as we are with the human developer experience. The primary interfaces that the agents will be interacting with, such as agent CLIs and SDKs and Skills, are things we need to lean into and invest in.”

In the video below, watch Barnett speak about agentic support in software development with Niv Raz, a senior solution architect with AWS, and Rob Kurver, a founding partner of the CPaaS Acceleration Alliance.

Vonage has already launched two MCP servers that expose its documentation and tooling to agentic AI. The MCP documentation server lets agents query it directly to receive context-aware answers. Since the responses come straight from official Vonage documentation, the results are accurate and allow for faster iteration.

The second server provides tools for managing Vonage applications and reports programmatically.

So if a developer asks an AI assistant to “show me how to send a WhatsApp message using Vonage APIs,” the AI grabs the relevant documentation directly from the documentation server. Then later, if the developer prompts the AI to “create a new Vonage app with messaging enabled,” the tooling server calls the right API to build and configure the app without any manual setup.

Additionally, Vonage has created a plugin for Kiro that brings Vonage's APIs directly into the Kiro editor. Kiro is a new coding environment from AWS that uses AI to help developers write and deploy software faster. It's the first communications platform to be natively available inside Kiro, giving Vonage an early-mover advantage with AWS's large developer base.

Expect agentic traffic to continue growing exponentially. Cisco predicts AI inference traffic will be 25% of total network traffic by 2035. 

Agents may one day have their own online identities, including logins, inboxes, and even software licenses, Microsoft executive Rajesh Jha said at a recent conference.

Microsoft is looking closely at what permissions to give AI agents within the company, CEO Satya Nadella said in a recent podcast discussion with LinkedIn Co-founder Reid Hoffman. “You need to give them identities, you need to give them sandboxes, then you need to set policies to govern them,” Nadella added.

Most software products still have no real way for an agent to register, even though agents are writing code, querying systems, updating records, and deciding which tools to use, wrote Michael Grinich, the founder of WorkOS, on LinkedIn. WorkOS has created an open protocol called Auth.md, which provides a standardized way for AI agents to register for web services.

“Agents are becoming real actors on the internet,” Grinich added. “They should not need to pretend to be humans. And users should not need to hand them master keys.”