Office work got automated, deskless work is next.

Office work got automated a long time ago. Tools like Zapier have been automating admin work for over 15 years now, so marketers, coders, customer service staff and other knowledge workers all have really good, proven automation options - but what about cleaners, baristas, plumbers and other deskless workers, where's their AI automation?

If your job happens in front of a keyboard, there are lots of great tools to make it faster by automating email inboxes, spreadsheets, accounts, shift scheduling, CRMs, code and so on, and since the arrival of AI agents, much of the admin work for these roles has been completely delegated to thinking machines.

It's been interesting for me to see, because over that 15-year period I was a consultant to deskless industries. Much of my work was behind the keyboard and I did benefit from those automations, but the people I was consulting for didn't have that luxury.

I started as a graduate manager in a builders merchant, learning how a team and a stockroom actually run at 8am on a Saturday when there's a queue of 25 vans beeping to get through the gates and be served. I then moved into health and safety on construction sites, running the training that kept people safe, before I built a consultancy and spent years inside retail and hospitality businesses, working next to the managers who open up, close down, and hold the whole thing together through thousands of daily mini tasks.

In every one of those places, the office had good software and the floor had almost none. The manager had a laptop for the numbers, but the deskless team had a WhatsApp group chat and a laminated checklist on the wall. Almost all the communication in these teams was checking whether someone had done something they were supposed to.

It's quite funny when I think back to some of our early solutions for this type of thing when I was a consultant. At one stage, we were creating these lovely branded management diaries, which we used to send out to hospitality and retail venues. We had handcrafted a huge set of tasks and activities that each venue needed to have done on a daily basis, packaged them up into wire-bound diaries and we used to post hundreds of those out every month.

It was a pretty crude but effective solution to one part of the problem - how to remind staff what needs doing on a daily basis and give them an easy way to log it. But there was absolutely zero automation involved, obviously. It did hardly anything to speed up a worker's day. It was mostly a management tool to make sure that everything was recorded correctly.

And by the way, let me just qualify this gap for you. Around 80% of the world works in a role that is at least sometimes away from a desk. They are the largest part of the workforce and they got the smallest part of the tools.

I don't think they have been the last to get automated because the work matters any less. I just think it's genuinely more difficult to automate work that changes constantly and is made up of thousands of mini pieces.

But let's be honest about the other half: the software industry went where the money was easy. Office workers built tools for office workers, and the largest workforce on earth got treated as an afterthought for twenty years.

Automating admin work is genuinely easier because the work's shape doesn't change as much. If you think about work in an office, the tasks are usually quite repetitive. For example, if an invoice comes in and needs processing, the system for doing that is generally very similar every time an invoice comes in.

But if you think about a task in a deskless team, it's often extremely variable from day to day, or even hour to hour. The times that things get done and the way things get done are far less predictable. So automating this deskless work has been skipped.

I think this is the moment it stops getting skipped. And I think, in a few years, we will look back and find it obvious. Running a deskless team on a WhatsApp group and a laminated sheet will soon look as reckless as doing the accounts on the back of an envelope.

Here are the two reasons why now is the time to automate deskless work:

1. Every deskless worker, or at least every deskless team, can carry around a tablet or phone which is basically a computer, thousands of times more powerful than the one that landed on the moon. These devices have become so cheap that the business cost is negligible.
2. AI agents have become much more reliable in the last 12 to 18 months. So much more reliable that they can now make decisions and take actions on more complex and variable data, i.e. the data that comes from deskless teams, which is not always of a uniform shape or standard.

These two things together are really important, because it means we can do two things for these teams. We can organise them at scale in a way we couldn't before, and secondly, we can embed AI into the work these teams carry out.

This sounds really obvious when you put them together in a sentence, but I think, genuinely, until very recently the future of work for these types of people was not obvious at all.

When AI chatbots were released a few years ago, it was pretty amazing to all of us, including me. I was still a consultant to deskless teams all the time, and it was really obvious that these chatbots could be embedded in their work and could help these teams, if the teams were able to access them and ask them questions.

But I also recognised really early on that this would be a really limiting way of using AI in these teams. It could certainly release some of the pressure on management workload, let's say, if AI could answer questions that staff had, but this type of solution was always going to live at the side of the work, accompanying it rather than being embedded in the work itself.

And that's still most of what gets sold as "AI for deskless teams" today: a chatbot bolted to the side of the work. Staff can ask it questions. That's a help desk, not automation, and it was never going to change how the work actually runs.

For the last 3 years, I've been obsessed with building a solution where AI can be embedded in the work, genuinely.

I wasn't satisfied with a solution where AI couldn't make decisions and take actions independently in a workflow. To me, it felt like if the AI couldn't do that, then it wasn't ready. These teams don't have time to be messing around, constantly setting things up and fiddling and tweaking. They need something which is build once, run properly forever.

The other thing I was really conscious about was the shape of the work that happens across these deskless teams. As I mentioned earlier, it's extremely variable from business to business, and even site to site. So giving users a fully customisable setup, so they could implement their own procedures, was an absolute non-negotiable while we've been building this.

The solution we came up with was a workflow builder, flexible enough for users to implement their processes in any way they want with a series of inputs, but also a way to actually embed AI decision making and actions into the workflow.

So this means we have created something where managers of deskless teams can organise their staff around workflows, i.e. the work to be done, but they can also organise agents around the same work. Not agents that are bolted off to the side or do separate jobs. I'm talking about agents that actually evaluate the work staff do as they do it, make decisions, and take actions based on the inputs and the things that staff log.

So how is this actually different from something like Zapier, or the other automation tools that have been around for years? It's a fair question, and it's worth nailing, because the difference is really the whole thing.

Zapier automates work that already lives inside software. Its entire job is to connect one app to another. An invoice lands in an inbox, so a row gets added to a spreadsheet, so a message gets posted in a channel. Every part of that happens between machines. Nobody has to walk anywhere, check anything or pick anything up. The work was already digital and the automation just moves it along faster. That's genuinely brilliant for admin work, and it's exactly why office roles have been so well served for the last 15 years.

But you can't Zapier a fridge temperature check. There's no app that the check lives in, because the check doesn't exist until a person walks over, at a certain time, and actually does it. And that's the whole problem with deskless work. The trigger isn't a piece of data arriving in a system, it's a human being, in a place, doing a physical thing - checking a fridge, inspecting a pallet before it ships, completing a care round. Traditional automation has nothing to grab hold of, because the work never touched a computer in the first place.

That's the gap we are filling. Think of it like this: Pilla is the thing that gets the work into the computer to begin with. The team runs the workflow on their mobile device, right where the work happens, and logs what they do as they do it. And then, and this is the important part, we put the AI inside that workflow rather than off to the side. So the moment a member of staff logs something, an agent can read it, make a decision and take an action off the back of it - chase the missed step, flag the failed check, escalate the problem, or update another app like Zapier does.

The apps these teams use today are clipboards with a login. They record that work happened. They do nothing when it doesn't.

So, back to the question I opened with: where's the AI automation for deskless workers? It's finally here. That's the whole reason Pilla exists.