Airbnb's AI playbook for HR (as told by their CHRO at Stanford)

A few days ago, a Stanford researcher published a summary of a talk that Airbnb's CHRO gave on how he's using AI inside his HR department.

I read it. Then I had to read it again.

This isn't consultancy theory. It isn't a future-of-work panel where nobody actually commits to anything.

It's the real playbook of one of the most recognised companies in the world, told by the person who's running it.

Here's what I want you to take from it.

Airbnb operates in more than 220 countries. It has tens of thousands of employees. And its platform is used by millions of people every month.

Iain Roberts is its CHRO.
He's been in the role for years.
And a few days ago he stood at the AI for Organisations conference at the Stanford Institute for Human-Centered AI and explained, in detail, exactly what he's been doing.

What makes this matter isn't the company. It's who's speaking.

Not a software vendor pitching you a solution. Not a consultant telling you what companies "ought to do". It's a sitting CHRO at a global company, with real context, real problems, and solutions that are already in production.

There's a huge difference between the two. And in HR we've spent years listening to far too much of the first kind.

His exact words: "Slides and PDFs are the enemy of intelligence."

Markdown is machine-readable, version-controllable, and verifiable. The same properties that made software useful for building things in the first place. PDFs have none of those three. They're a snapshot. A frozen document the AI has to guess at, not read.

Airbnb's leadership team already writes in markdown. It isn't an aesthetic detail. It's an infrastructure decision: every strategy document, every policy, every meeting note now lives in a format the AI can actually work with - search across, version-compare, spot contradictions in, summarise without making things up.

My read: if your HR documentation lives in PDF and PowerPoint, your AI can't work with it properly. This is the step zero almost nobody takes - and the one that conditions everything else. Before you even start thinking about agents, automations and copilots, ask yourself: what format does your department's knowledge live in? Because if the answer is "PDFs in Drive folders", you don't have a tooling problem. You have a foundations problem.

Airbnb's HR team wasn't known for executive storytelling.
The talent was uneven: some people communicated brilliantly, others didn't. And that meant good ideas were getting lost in mediocre presentations.

What they did: they trained a model on Airbnb's best executive presentations - the ones that actually worked, the ones that moved decisions - and turned it into a skill that anyone on the HR team can use. You drop in your data, you get a polished narrative in seconds, with the structure and tone you already know works inside your organisation.

The result isn't that everyone writes the same way. It's that nobody starts from a blank page, and the quality floor lifts for the whole team.

My read: this isn't "go and use ChatGPT". It's building a reusable asset out of your own company's knowledge. The difference between using AI and owning AI. One is a recurring expense that produces generic output. The other is an investment that captures what your organisation already does well and puts it within reach of everyone. A well-built skill pays for itself every week, with every person who uses it.

Airbnb ingests video from every town hall - VP through C-suite - and measures two specific things.

First: how long it takes for a message to travel from "this is where we're going as a company" to "this is what it means for my work". How many days, or weeks, pass between the CEO saying something at an all-hands and a team manager translating it into a concrete priority for their people. That's cascade speed.

Second: where there's dissonance between the strategy that's being stated at the top and what an area is actually doing day to day. If leadership says "this year we're prioritising profitability" but a department keeps talking only about growth in its meetings, there's a crack. That's communication accuracy.

My read: this is real people analytics. Not the annual engagement survey that takes three months to process and measures what people say they feel, rather than what's actually happening. This is a real-time signal telling you whether the strategy is landing or getting lost on the way. We've spent years in HR wanting this data and settling for proxies. The technology to have it for real already exists.

His team handed him a 6-month roadmap for an offsite-planning tool for hybrid teams.

He built it himself over a weekend.

His line: "If you delegate the building, you delegate the learning."

He didn't do it to save six months. He did it to understand, hands-on, what can and can't be done with these tools today. And then he made his leaders build things with AI - not supervise others doing the building, but do it themselves - so they learn and experiment alongside their teams, not above them.

My read: this is the most uncomfortable and the most important of the five. The CHRO who only approves the AI budget but has never built anything doesn't understand what they're approving. They can't. It's like running a product team without ever having used the product. Building isn't optional for anyone who wants to lead this transformation. You don't have to become an engineer. But you do have to have had the conversation with the AI, hit a wall with it, iterated. It's the only way to genuinely understand what we're talking about.

Palantir popularised the Forward Deployed Engineer model: someone who connects the data, maps the process, plugs in the AI and moves on. Functional. Technical. But limited - it treats the organisation as a system of pipes, when it's something else.

Iain Roberts thinks that falls short.

It isn't enough to bolt AI onto the processes you already have. You have to redesign the organisation around people and technology together, not in parallel. And for that you need a different kind of profile. His exact description: "An embedded organisational architect. Someone deeply empathic, deeply social, who understands the human condition - that an organisation is really a network of individuals trying to get things done together."

My read: that profile is exactly the new VP of People. Not a technical person. Someone who understands people AND understands AI, and who redesigns the organisation with both at the same time. The FDE optimises what already exists. The organisational architect redesigns what ought to exist. It isn't a role that's coming. It's the role that companies who are serious are already looking for - they just mostly don't know yet what it's called or where to find it.

These aren't five disconnected tools. They're five decisions about where AI lives inside your HR function.

Airbnb's CHRO isn't waiting for IT to bring him AI. He's building it himself, with his team, redesigning how HR works from the inside.

That's the difference between leading the transformation and receiving it.

And that difference is what's going to separate the CHROs who matter from the ones who don't over the next two or three years.

Audit what format your HR documentation lives in. If it's PDF and PowerPoint, that's your step zero. Start there - before reaching for any AI tool.

The "embedded organisational architect" Iain Roberts talks about isn't science fiction.

It's the profile we're building inside the AI VP of People Fractional Programme at Próximo, which has just kicked off. Not a technical person. Someone who understands people AND redesigns the organisation with AI, from the inside.

If you want to build that role inside your company - exactly the one Airbnb's CHRO describes - reply to this email and I'll tell you whether it's the right fit for you for future cohorts.

That's what this newsletter is for. Every week I tell you what's happening, the cases that matter, and how to get ready without waiting for someone external to explain it to you.

Stick around. I'll keep you up to speed.