I'm not a typical AI user: I experiment constantly. A more interesting question is how AI-native companies are using AI today. This is what I'm seeing in the leading companies I speak to and that we work with.
1. They measure business results, not AI activity
Nobody gets credit just for building an agent or an app. What matters is whether quotes go out faster, customers get better answers or employees spend less time on admin. They measure quality and cost too, including the time spent checking and fixing the output. Automations that don't help get changed or stopped. Token maxing doesn't exist.
2. All systems are connected
One of the first things AI-native companies do is connect all relevant data, tools and systems to their AI platform of choice: CRM, ERP, HR systems, expense management, advertising, e-commerce, payments, customer service. You name it. This gives every employee access to the information and tools they need.
Thanks to progress in MCP and in the tool calling of LLMs, these connections can often be made in less than a week for a small or medium-sized organization.
3. A sandbox to experiment and build
They give essentially everyone in the company a safe place to connect to internal (and external) systems and to build agents and automations. In regulated industries, the company makes sure the system detects and alerts when a use of AI is non-compliant. This reduces the need for restrictions and lengthy policies (see point 7).
4. Deploying to production is fast and frictionless
Any employee can contribute agents and automations, which go to production as soon as they have been tested to be useful, secure and maintainable. For example, an accountant in finance comes up with an agent that improves expense reporting, and after a one-month pilot it goes into production for the entire company.
Anthropic is the ultimate example of this, launching new products in a matter of weeks.
5. Workflows are reinvented, not automated
Don't turn existing workflows into AI automations as they are, unless they are a perfect fit for AI. Instead, let employees, customers and other stakeholders dream up how they would like things to work. For example, the sales team never has to log in to the CRM again unless they want to: their agents update customer information, leads and opportunities automatically from emails, meetings and other communication.
6. Everything is shared by default
In most companies, 95% of LLM interactions are private sessions. In AI-native organizations, 95% are shared: chats, agents, automations. This means that:
- people learn from each other;
- when someone leaves, their content and agents are still there;
- the organization builds up knowledge and memory over time.
7. A simple and clear AI policy
The purpose is not to use AI, or to build agents and software with AI. The real goal is to achieve the mission, vision and goals of the company, with AI as one enabler. Don't replace what already works: if the company has a CRM that works, don't replace it to save $1,000 a month in licence costs.
8. They build feedback and testing into the workflow
An agent working once is not enough. They test against real examples, record failures and use employee and customer feedback to improve it. When a model, a prompt or an integration changes, they check that the workflow still works.
9. Every production workflow has an owner
Employees can experiment, but shared agents don't become ownerless. Someone is responsible for performance, permissions, maintenance and what happens when it fails. The team knows how to pause it and carry on without it.
10. Humans have defined responsibilities
Decisions and tasks that humans do better than AI (or where AI doesn't deliver quality output) are done by humans. Workflows include human approvals and judgement, and the AI policy makes this clear.
In short
AI-native companies don't just give everyone AI tools. They make useful work easier to repeat, improve and share, without losing accountability.
If you want to discuss how your organization can become AI-native, or move further along the path, feel free to reach out.