This deep dive expands on the Loop Asia conversation with Aki Ranin, a two-time founder and AI advisor who co-created the “Usable AI” transformation framework.

Stack Overflow's users started deleting their own accounts in protest — and Stack Overflow moved to stop them, because the data itself was the asset being sold. That's the sharpest illustration of a question I put to Aki Ranin partway through the episode: if an AI agent goes out to find a service or a data source on a business's behalf, how does that business make sure it gets found, and found correctly? For twenty-five years the equivalent question had a known answer — SEO, good metadata, a sitemap Google could crawl. Ranin's answer made clear this version doesn't have a settled playbook yet — and the businesses that work it out early get a real advantage.

Two roads, and most companies haven't picked one

Ranin laid out the landscape plainly: businesses can block AI crawlers from OpenAI, Anthropic, and others outright, or strike commercial deals — the way OpenAI did with Reddit — buying access to content the platform owns but the users who created it never really controlled. Stack Overflow saw the same dynamic from the other direction, monetising contributions until its own users revolted.

His broader point lands on something most content and marketing teams don't say out loud: "whether it's Google or Instagram, Facebook, et cetera, you own none of the content." The terms and conditions have always said this — it just didn't matter when the only thing being extracted was ad targeting. Now that the same content can be repackaged as training data or a citation source for an AI answer engine, the ownership question has commercial teeth it didn't have five years ago.

For a COO or Head of Professional Services — anyone whose mandate includes how the company shows up in front of prospects — this decision deserves the same deliberateness as any other channel strategy. Do you want your case studies and technical documentation crawled and cited by the tools your prospects now use for vendor research? Blocking crawlers by default, out of inertia, may quietly remove you from consideration in exactly the venue where buying research increasingly happens.

What "documentation" means when the reader is a model

Ranin's assessment: for API documentation, the problem is close to solved. Modern models produce reliable structured JSON, so reading existing documentation and generating clean documentation from a codebase are both now tractable. If yours is thin or trapped in a wiki nobody maintains, that's a fixable gap, cheaper to fix now than ever.

The more interesting question: what happens once documentation stops being the bottleneck? His instinct is that consumption speed will matter more than structure — video with an audio transcript could out-perform a JavaScript-rendered website, because a model reads transcript text immediately while scroll-triggered loading makes it wait on a pattern built for human convenience. Carried to its extreme, the fastest path is direct database access — risky in most real cases, he noted, but the direction of travel is the point: every layer of human-oriented interface between your data and the agent reaching for it is friction a machine reader doesn't need.

The strategic version of this question

Put the two halves together and you get a genuinely new question: what is your business's machine-legible surface, built by accident or on purpose? Most documentation and case studies were built for human readers navigating a browser, never audited for what a model answering "does this vendor do X" actually needs — clean structured answers, not marketing copy wrapped around them.

This is integration and API strategy work aimed at a new class of consumer, sitting at the boundary between content and engineering — which is why it falls through the cracks in most organisations. Getting it right means treating "can an AI agent correctly understand and use what we expose" as a design requirement alongside the human-facing site. It's the same question Blue Connector puts to a client before anything ships: not whether documentation exists, but whether a model reading it gets a straight answer or a marketing paragraph.