One's Business Brief: The AI Worker

The history of technology is also the story of how work gets done. From hand tilling to plough; plough to tractor. We develop novel ways of completing work with less energy—or by channelling stored energy through machines and devices instead of our own bodies.

The same goes for the work we do in our heads. Knowledge work now constitutes the majority of the workforce in high-income countries (60% in the US). And just as our ancestors innovated the ox-drawn plough, we developed software to outsource our thinking work to computers.

Software, from the outset, was a complicated thing to build though. You needed specialists to develop these systems. To build up libraries. To innovate new programming languages with higher levels of abstraction.

But at its core, building software has always been a matter of telling computers the work we want them to do. A way of outsourcing our thinking work to processors in a device rather than the neural networks in our heads.

The vast majority of software users are not developers though. They are dependent on developers to build the tools, features and interfaces they need to get computers to help them with their work. So, it was only natural that platforms would arise to help anyone build their own software—albeit with limited functionality—not just developers.

Notion, Webflow, Zapier were the no-code poster children of a new world where anyone can build their own software and outsource more of their work to computers. You didn’t need to go into an IDE and write Javascript or Python to converse with your computer—you could just use lego-like building blocks to construct your own solutions. Lovely.

However, as no-code continued to progress, it became clear that there were limitations to the work it was capable of doing. A lot of what no-code tools seemed to offer, organically, might be described as organizing work, rather than actually doing the work.

Perhaps I’m nit-picking, and organizing work is of course itself a form of work. Sorting contacts into lists, filtering information, sending information from one app to another, automating a pre-written email when an event occurs. These are all, of course, forms of work—but they are somewhat limited to moving information around, rather than processing and even generating new bits.

Enter AI. We’ve all had the experience of watching ChatGPT spin up 500 words in response to some question or prompt or idea, and felt in a very real sense that it was generating something new, on the spot.

This is the promise of AI. That we’ve innovated a technology which is capable of doing more of the knowledge work we face every day, rather than just aiding us in preparing and getting ready to do the work ourselves.

And just as the first generation of no-code platforms allowed anyone—developer or otherwise—to build their own little solutions for organizing work, we will need platforms that are capable of letting anyone build their own AI workers.

For me, an AI worker has the following characteristics:

• It is responsible for completing a concrete job (often very narrow in scope);

• It has a unique knowledge base and context to work from;

• It has clear instructions for the work to be done;

• It has a history which it can learn from (feedback about prior jobs, etc.);

• It has access to a suite of functional tools which it can decide to access depending on the job at hand;

• It has an ability to make decisions based on variable inputs; and

• It can meaningfully process information to produce lower entropy results.

An AI worker is not an AI employee, in the way you might see headlines talking about AI replacing employees. It doesn’t work on broad, long-term projects, and it has very limited autonomy.

An AI worker is not an ‘AI agent’, either. This is a term being used in a very specific research context, to describe AI-driven systems which have an internal loop that persists until the job is completed.

No, AI workers are more like simple worker drones. They have a very narrow and focused task to do, so they keep their heads down, power through, and report back when they’re done.

But unlike prior solutions (workflow automations, etc.), they are built with an AI-first approach, with an appreciation for the new flexibility and general knowledge work capabilities of the latest LLMs. You don’t need to statically define every single step of the process, rather you can give a rough outline of instructions and let your AI worker fill in the gaps.

Instead of trying to fit software development paradigms on top of a no-code interface (e.g. mapping the fields of a JSON object one-by-one with fields in a form), we can now leverage the intelligence of AI models and build interfaces that work in a fundamentally different way.

I see AI workers being an important next stage of software development. They will be responsible for an enormous amount of the knowledge work that gets done, every day, in literally millions of businesses around the world.

There will be complex, vertical AI workers built with custom software, by developers. And there will be a parallel no-code development framework, too. This is what I’m hoping to contribute, through Bizway; a no-code platform for building your own AI workers.

I can hardly say Bizway is there yet, but I think it’s helpful for you, as a user, to know the direction I’m heading. The larger vision is that anyone will be able to build and manage their own little fleet of AI workers to help them complete hundreds of discrete tasks in their business every day. And to host and manage them all in Bizway.

Lots of work to be done between now and then, but that’s where I think this is all heading.

If that sounds exciting to you, let me know. If it worries you, let me know why. I’d love to hear your thoughts.

Gerrard
Founder, Bizway.io 

PS - I’ve now updated to wording across the Bizway platform to be ‘AI Worker’ focused, so Workflows are now Workers, etc. If you’d like to build your first AI worker, you can still do so by trying Bizway with all features, free for 7 days.