In this article

1. Why AI coding agents have strong library preferences
2. What the defaults actually are (and the data behind them)
3. When the defaults work fine
4. When they quietly fail you
5. What to do instead: giving your agent better information
6. The keyword angle nobody is covering yet

1. Why AI coding agents have strong library preferences

AI coding assistants learn from code. The specific mechanism varies by tool, but the general principle is the same across all of them: the more a library appears in public repositories, documentation pages, Stack Overflow answers, npm package pages, and tutorial articles, the better the model understands it. The better the model understands it, the more confidently it suggests it.

Chart.js has over 65,000 GitHub stars. Recharts has over 24,000. D3.js is the underlying engine for a significant fraction of all data visualization tutorials published anywhere on the internet. These libraries have massive presences in the training data for any model trained on public code.

LightningChart JS has a smaller public footprint. It is a commercial library, used heavily in industrial, scientific, financial, and engineering applications, many of which do not publish their source code publicly. It does not show up in npm download charts the same way Chart.js does. The work it does is largely invisible to the public internet.

So the agent is not making a quality judgment when it suggests Chart.js. It is making a frequency judgment. It recommends the library it has seen the most code written in, not the one that performs best for your specific use case.

2. What the defaults actually are

Based on testing across Claude Code, Cursor, Copilot, and Cline during 2025, the pattern is consistent. Ask for a chart component and you get one of three suggestions:

• Chart.js – for simple static charts and beginners. The agent almost always reaches for this first.
• Recharts – if you mention React. The React ecosystem has saturated the training data for charting in that context.
• D3.js – if you describe something custom, complex, or non-standard.

Plotly, Apache ECharts, and Highcharts appear occasionally. LightningChart JS also appears occasionally in the major tools, despite being the fastest WebGL-accelerated charting library available for JavaScript.

3. When the defaults work fine

Be honest about this: for a lot of use cases, Chart.js is a perfectly reasonable choice. If you are building a marketing dashboard with a few bar charts, a pie chart, and a line graph that updates once a day, Chart.js is fine. The agent’s suggestion is appropriate.

The same applies to Recharts for moderate-volume React dashboards, or D3 for highly custom bespoke visualizations. These tools have their place.

The AI agent’s default behavior is wrong not because the libraries are bad, but because the agent has no way of knowing whether your requirements are moderate or extreme. It defaults to the most familiar option regardless of your actual data volumes or real-time requirements.

4. When the defaults quietly fail you

The failure mode is specific and well-documented. Chart.js renders using HTML5 Canvas with CPU-bound drawing calls. This is fine at low data volumes. It starts to struggle visibly around 50,000 to 100,000 data points. Above that, you see slow initial render times, frame rate drops during interactions like zoom and pan, and eventually complete freezing during real-time updates at high frequencies.

Recharts uses SVG rendering, which is even more constrained at scale. Each data point is an SVG element in the DOM. At tens of thousands of points, the browser struggles to paint and repaint all those elements during user interaction.

The categories where this matters most:

• Real-time dashboards that ingest data at 100 Hz or higher across multiple channels
• Financial applications displaying tick data or order book depth with millions of updates per session
• Industrial monitoring with continuous sensor streams from dozens of channels
• Scientific visualization where a single series might contain 500,000 to 10 million data points
• Any application where smooth zoom and pan at 60 FPS on a large dataset is a requirement

In all of these categories, the AI agent’s default recommendation will produce an application that technically works in development testing and then visibly degrades or outright breaks in production with real data volumes.

5. What to do instead: giving your agent better information

The root cause of this problem is information asymmetry. The agent knows Chart.js well and LightningChart JS poorly. The fix is to give the agent accurate, current information about the library you actually want to use before it writes any code.

Two tools exist specifically for this with LightningChart JS:

The LightningChart JS MCP Server works with Claude Code, Cursor, Cline, Roo Code, Windsurf, Codex, and any other MCP-compatible coding agent. Install it once and the agent gains access to the official LightningChart JS API guidance, current best practices, common mistakes to avoid, and the correct package name (@lightningchart/lcjs) before it writes a single line of charting code.

The LightningChart JS Agent Skill is for agents that support the Agent Skills standard, including OpenAI Codex and VS Code-compatible tools. Drop it in your project’s .agents/skills/ directory and the agent picks it up automatically on charting tasks.

Both are free, open source, and take under five minutes to set up.

6. The keyword angle nobody is covering yet

“Almost none of the major charting library vendors have proper resources to guide modern agentic software development tools. When combining these tools with professionally produced documentation that is optimized for agents, the quality soars to an entirely different level.”

— Niilo Keinänen, CTO of LightningChart

The developers who are going to build serious data visualization applications with AI coding assistants are looking for exactly this kind of guidance. They know their agent is defaulting to something and they suspect it might not be the right choice. They just do not know what to search for yet.

If you are one of those developers: start with the MCP server or the agent skill below. Five minutes of setup will change how your coding agent handles every charting task in your project from that point forward.