Why Narrative runs on headless software

Business software is moving towards being headless: used by an AI on your behalf, with people logging in far less often.

At Narrative we have switched to tools that support this. Attio is our system of record for customer data, but we rarely log in to it, because our AI works with it directly. Documents live in Notion, where our AI can read and organise them. Email runs through Superhuman, and we use Slack, Mintlify for product documentation, Braintrust for evaluations and Linear for product development, all connected to the same AI.

What makes software good to work with this way

Access to the data is only part of it. Good headless software also defines the rules for working with it. If there are fifty things you can do in a CRM, the best products describe each of those workflows in a way an AI can follow reliably.

The benchmark to watch

AutomationBench, from Zapier, tests exactly this. It places an AI in an organisation with many systems, gives it a task and asks it to work out how to use those systems to complete it. In early September the best models completed around 30% of its tasks, and GPT-6 has since raised that to about 40%.

Why it matters for banks

The same shift will reach banking software. Narrative Coach is built to work this way: it joins a bank’s app through an SDK, connects to its CRM and is documented so that both people and AI tools can follow it.