Our AI already does the work. Next, it owns the whole job.

Everyone is worried AI will take their job. AI can now do almost everything a human can do at a computer. I think that’s true.

But a job is two things. A list of tasks. And a list of responsibilities.

AI already has the first half. The second half is the hard part. Even if AI does every task, someone still has to own the outcome. When something doesn’t happen, you want to be able to go to someone and ask why. So far, that has always been a human.

Closing that gap is what we’ve been working on at All Gravy for the last year and a half. Every step has moved AI a little further away from answering and a little closer to owning. Here’s where we’ve been, what changed along the way, and what it means for the people running HR in hospitality. (It’s also an episode of our podcast.)

Step one: it answers

We started with the obvious low-hanging fruit. Frontline staff need answers, fast. How do I call in sick? What’s the allergen process? Where’s the policy on breaks?

So the first thing our AI did was find the answer in your own handbooks and give it back with the reference. Useful. But it only answers.

Step two: it does the work

Then we saw what AI agents were doing inside our own company. And we thought: if we can put that power to work for our customers, that’s huge.

So we gave an agent access to every tool, every workflow and all the information inside All Gravy. That’s what our AI Assistant is today. It can do pretty much any job you would do in the platform. Onboard new starters. Build a course and a quiz from a policy. Write the update and send it to the right sites. Find who’s behind on training. Turn a photo of a paper rota into a rules-checked draft.

It’s been a big success. People with access get around 60 tasks done with it every month, on average.

But we noticed two things.

Some people get loads of value from it. Others never discover half of what it can do. It’s all left up to the user.

And it takes active thinking. You have to go in and ask. That turned out to be the limit on how much value it could deliver.

Then AI moved into the group chat

The first fix for that limit didn’t come from us. It came from how everyone started using AI.

Meta has started putting AI agents straight into WhatsApp and Messenger. Ask one to book theatre tickets or plan a whole trip, and it just does it. In the same thread you use to talk to your friends.

That’s a big deal. Not because of what the agent can do. Because of where it lives.

When AI was for the nerds, the tools showed you everything. The reasoning, the calculations, every step on screen. Hard to follow unless you were already deep into it. Now it’s a normal conversation. “I want to go to the cinema with my kid.” “Here are three films showing near you. Want me to book tickets?” AI feels like another person in your chats.

We got lucky with where we built. All Gravy was made to feel like Facebook and WhatsApp for your business. Feed, chat, groups. So our AI sits right in the chat, next to your colleagues. You message it like you’d message anyone else on the team.

That changes who uses it. Before, working with our AI felt like the advanced tools power users already live in. Great for HQ people. Less great for a General Manager in the middle of a Friday service. Now the GM can work with it too. It’s a teammate, not a complicated piece of software.

The younger generation skipped the debate entirely. At the start, frontline staff tried to break the AI. There was a fair bit of “screw AI” in the air. Lately it’s the opposite. Instead of clicking through the onboarding journey we set up, some new starters just ask: “What do I need to know? What matters most for me?” They onboard themselves through a conversation. Dialogue first, menus second.

But a teammate you still have to ask is only half a teammate. Which brings us back to the missing half of the job.

An agent has agency

There’s a lot of talk about the difference between an agent and an automation. A slightly boring topic. People love it anyway.

My answer is simple: an agent has agency. It does things on its own accord, the same way your colleagues do. Put the bar there and suddenly most things called “agents” don’t clear it.

Clearing it takes two steps.

First, it works before you ask. That’s what we’re building now. Say you’re Head of Compliance. Every morning the agent checks the compliance reports, tells you how things look today and flags anything that needs you. You just read a digest. Or it watches training at every location, spots a site that’s dropping and escalates the people who aren’t getting through.

Then, it owns an area. Not a task. An outcome. “Make sure every new employee gets properly onboarded” covers the contract, the paperwork, the training and the first-day messages. An agent that owns that area is responsible for all of it. If a new starter falls through the cracks, the agent has failed.

That’s what has made agents hard to use so far, for me personally too. You have to manage them all the time. They execute your tasks. They don’t take responsibility for a job. An agent that owns the job flips that. Less time telling it what to do. More time on your own work.

That turns better software into an actual virtual employee. You’ll see them in your employee directory. You’ll see them in your chat. You talk to them like anyone else on the team.

Meet Emma

The first one is our onboarding agent. We call her Emma.

Emma owns everything from the moment you decide to hire someone to the day they’re fully ramped:

• Sends the contract
• Does the right-to-work check
• Lets the manager know when it’s all done
• Puts the new starter on their first three shifts, with their manager
• Reminds the manager the day before: your new starter is in tomorrow

And when something goes wrong, she deals with it. Missing data for the right-to-work check? She goes back to the new starter and asks for it. Still nothing? She escalates to HR. The new starter gets through either way.

It’s very similar to hiring a new teammate.

Think about the HR person who spends all day chasing contracts and right-to-work checks. Creating people in systems. Taking them out again when they leave. Calling managers to ask how the new person is doing. None of that is the work they actually want to do. So they hire an HR coordinator to take it off their plate.

Emma is that coordinator. She introduces herself like any other colleague. “Hi, I’m Emma. These are the jobs I’ll do for you. Tell me if there’s a way you like things done.”

So you tell her. “Two days before someone starts, their manager sends them a message. This is the feeling we want them to have. This is how we talk to new people.”

Emma learns that. Then she gets on with it.

What’s nice about Emma is that she does the same thing every time. You know the right-to-work check gets done. You know the system gets updated. But she also has the agency to act when things don’t go down the happy path. And you always have full visibility: open the dashboard, see who’s in the onboarding flow, tell Emma to do something or take over yourself.

How much should you stay in the loop?

People still hesitate to fully trust AI. Fair enough. But part of that hesitation was the old interface. When an agent narrates every step (“now I’m checking this, now I’m looking up that”), it feels like something you need to babysit.

When it just says “I’ll do that for you” and comes back when it’s done, trust builds slowly. The same way it does with a new hire.

How much you want to be in the loop depends on the job:

• Changing your onboarding for new employees? You want to be involved. Explain the why, look at what it builds, give feedback, iterate.
• Sending a contract to a new starter and getting them set up? You probably don’t need to be involved, unless something unexpected happens.

That’s why, today, our AI acts after a manager approves. You decide where that line sits, and you move it as trust builds.

Start where failing is expensive

Which jobs should go to agents first? The tedious, repetitive ones with a really high cost of getting wrong.

Layer 1: admin that can’t fail. Onboarding, with all the compliance and paperwork that comes with it. Compliance training, so people have done everything they need before they get scheduled. Payroll checks, so everything is approved and lives up to labour law.

Layer 2: work that adds value. Recruiting. Building learning and development content. People analytics.

Layer 3: strategy.

You move up through the layers. That’s the order we’re building in too. After Emma comes a compliance agent that looks after your whole business being compliant: training done on time, locations that are slipping, reports ready for your management meeting. If you use our Operations module, it checks the to-do lists too. Did someone check the fridge temperatures today? So you’re always audit-ready. A recruiting agent and a scheduling agent are in the works as well.

Right now we’re working with our customers to answer one question: which jobs do you actually want an agent to own?

What accountants learned from Excel

So what happens to the people whose tasks the agents take over?

The grunt work you do feels like your work. But really, it’s what you have to do in order to do your job.

Think about accountants when Excel came out. “I’m not going to calculate on paper and archive all these files anymore. Do we even need accountants?”

Accountants exploded. The grunt work got fast, so the value of the job went way up. They moved into financial planning and analysis. Not just what is the budget, but how are we doing against it.

I think we’ll see the same thing in HR and Ops in hospitality. It will feel strange at first. Not being the one who opens the spreadsheet and ticks off: training done, right-to-work check done. That’s just done for you now.

What the Head of People does in 12 months

Picture a Head of People a year from now, running a handful of agents. What are they actually doing?

They become system builders. They tweak. They make things better.

In hospitality, what you actually want from HR is simple. Attract people. Train them. Spot the next leaders. Keep them. That’s the work. But HR ends up doing coordination instead.

Free them from it, and the questions change. Instead of checking that every right-to-work check is done:

• What does the best possible first week look like? What if new starters got a welcome video on day minus one instead of day one?
• What’s our employer brand? What’s our value proposition to new people?
• What’s our retention rate by location? Which sites are doing worse, and why? What can they learn from the managers who are doing it well?

They iterate with the agents: “Let’s test this next time we onboard someone.” And the agents flag problems themselves. “Churn is high in the first three months. Something in the onboarding flow isn’t working.”

That last part matters most. The people side of these businesses is hugely important, but the data has always been missing. Without it, it’s hard to take that seat in the management team and the boardroom. (Our Hospitality Retention Benchmark 2026 is a good place to start.)

The HR manager becomes the decision maker and the strategist.

Nobody got into HR to do right-to-work checks

L&D teams went through this a few years back. Many wanted to be called “performance” instead. The point was to go from producing content to having an impact.

The same thing is now happening to every people role. When the admin goes away, you add to the bottom line some other way. Better training. A stronger employer brand. Attracting the best people. Spotting the best people you already have and promoting them.

And if you’re the person who built a system of five or eight agents running onboarding and training, you’re not at risk. You’re very employable.

AI will take your current job. That’s a good thing. You get to find a new one inside your role.

Nobody started in HR or People and Culture to do right-to-work checks. They started because they wanted to have an impact, be around people and give something to other people.

That’s what this makes room for.

Where we are today: the AI Assistant does the work when you ask, and acts after a manager approves. Emma and the other agents are what we’re building next, together with our customers.