What an AI-Native Platform Really Means

When I attended Startup Oasis in Atlanta last month, nearly every pitch was for a company, product, or platform that was “AI-Native.” And, just like saying “milk” 100 times makes the word strange and meaningless, by the end of the night the term had become unmoored from reality. Yet my company and I are part of the problem: Workmind is an “AI-Native” platform. So before I throw too many stones, let me shore up my glass house by defining what an “AI-Native” platform is.

“AI Native” in my own words

A company, platform, or product is “AI Native” when its designers have built it from the ground-up to be easily understood and used by Large Language Models (LLMs) and their harnesses. In my opinion, a platform can achieve this by offering three things to its AI users:

• Token Efficiency
• Discoverability
• Resumability

Token Efficiency

Today’s LLMs can only remember a limited amount of information, called a context window. The more information an LLM has to remember, the slower and more expensive it gets to run. That’s why the best products give AI users a way to get their information in a dense and specific format. While humans best understand visual information like an HTML web page, an AI agent prefers its information from a Web API where it gets text without any visual elements. An AI-Native product makes an API like this available to AI users, and keeps that API updated alongside its other product features.

Discoverability

It’s not enough, though, to simply make an API available – the AI users need to know that it exists and how to use it. This is discoverability, and an AI-Native product can optimize its discoverability in a few ways:

• llms.txt In September 2024, a consortium of AI companies proposed a standard llms.txt file websites can host to tell AI users how best to use the site or product. Major players like OpenAI, Anthropic, and Gemini all publish llms.txt files, and your product or website can, too.
• OpenAPI Spec Another, more technical standard is the OpenAPI Spec. While an llms.txt may be a text document instructing how to use an entire website or product, an OpenAPI spec is a structured document that is only about a product’s API. It describes the specifics for each tool available, what data they need, and what data they return.
• Text-based Documentation An AI-Native website or product not only publishes its documentation, but it is text-based and easily accessible to AI agents. Large Language Models take more time, tokens, and money to understand images than text, so the best documentation eschews images and fancy formatting in favor of concise, text-based content. Some companies, like Temporal offer both human-readable HTML files and LLM-readable markdown files for all of their help pages.

Resumability

Because of their limited context window, AI users have poor long-term memory, if they have memory at all. This means that every time an AI user does a task, it is like they are doing it for the first time; however, our databases and websites do have memories, and it is not often that a task starts from a blank slate. An AI-Native platform should remember important information on behalf of the agent, and share this information with the AI early in its interactions. For example, if an AI agent is trying to add a new contact to a list of leads, an AI-Native CRM checks if the contact already exists before creating a duplicate.

A platform could also offer summaries or accessible analytics about ongoing projects. These can help the AI user resume their prior tasks, or use past projects as a template or how-to guide for their next task.

GEO: Generative Engine Optimization

The last piece of an AI-Native platform is about a broader type of discoverability: GEO, or Generative Engine Optimization. Like Search Engine Optimization (SEO), GEO focuses on getting your product or website noticed by information aggregators and web scrapers. But GEO has a particular focus on appealing to commercial agents like ChatGPT, Claude, and Gemini. GEO is still a new field, and best practices are still emerging. But the main themes have started to coalesce around helpful, non-commodity content that is organized well with headers, navigation links, and is made available to web crawlers. Backlinks are also important to help AI bots discover the full extent of your website or product’s capabilities.

Why bother going AI Native?

Creating APIs and documentation for your website or product is not free and it takes time, so what makes it worthwhile? According to Cloudflare, one of the internet’s largest infrastructure providers, bots like AI agents make up about 34 - 38% of all web traffic today. That number will only grow as commercial LLM products like ChatGPT Work and Claude CoWork grow in popularity. Just like how a Facebook or Instagram page became critical to a business’s digital footprint, these AI-Native features will become critical to upcoming brands and products. And just as early adopters of social media platforms enjoyed a first-mover advantage, today’s AI-Native products can gain an edge with tomorrow’s growing AI audience.

Put AI to work across your systems

Exploring how AI agents could support your operations? Talk with Workmind about coordinating work across your connected systems.