Why NZ businesses are stuck in single-player AI, and what multiplayer looks like

We talk about this daily at work.

AI has two modes. And most businesses are only playing the one that has a very low ceiling, and is a central case for the limited ROI organisations accrue from AI.

Here is the pattern I see on almost every first call now with a prospect.

Problem is, the business still runs the way it always ran, just slightly faster in a few places.

McKinsey's research this year put a number on the same thing, with around 90% of businesses experimenting with AI and fewer than 10% capturing real value from it. That gap is not a technology gap, it is down to organisational approach, concretely it is the gap between the two modes. Very few businesses are clocking how to crack the second mode.

The first mode is single-player: one person, one chat window, faster at their own task.

The second is multiplayer mode: work is redesigned so one person's output becomes the next person's starting point.

In single-player, nobody inherits anyone else's progress. In multiplayer, everyone does.

I've touched on this earlier in the mistakes businesses make implementing Claude, where building single-player tools instead of multiplayer systems was one of the most common failures I see.

It also lines up with the AI Everest model I published last year, where Base camp and camp one of that climb are single-player by design: leadership alignment first, then individual fluency. Everything from camp three upward is multiplayer.

This article is about the line between those camps and why so few businesses cross it.

One thing to call out upfront: this topic is difficult to define and explain as to why some organisations cross into the second mode while most stay stuck in the first. What follows is the clearest map I can draw of the two modes, the crossing, and why it is hard.

**Single-player is easy to find.**Open the chat, ask, get an answer. Set up an individual Agent, and you're off.

It has an easy entry point and clear shape that one person can recognise, so everyone discovers it on their own, and the acceleration in work shows up very quickly in real work. This is the correct place to start. Individual fluency comes first, people build comfort, the reports get more accurate, and emails are faster to write. Nobody should skip that phase, it's a ticket to the game for every business. And, importantly, the failure is not being at single-player. The failure is not knowing there is anywhere else to go and figuring out how to get there.

Multiplayer mode has no obvious shape. Nobody stumbles into it.

It has to be designed, and here is the that makes it harder than it should be is that the tools themselves are built inherently single-player. They are sold per seat, chats are private and isolated to the person, and memory is personal to each user. So multiplayer is not a setting anyone can switch on. It runs against the grain of how the tools arrive, how we're taught to learn about AI, and that means someone has to have a strategy and build multiplayer mode on purpose.

This is the distinction I call Level 1 and Level 2. Level 1 is individual productivity, one person getting quicker. Level 2 is system transformation, the business itself working differently. Most companies are running a Level 1 rollout and reporting it as Level 2.

They aren't lying. They just simply have no picture of what Level 2 looks like, the benefits of it, and so they assume faster individuals is the destination. Unfortunately, that is where the ceiling hits.

The key problem with single-player mode is that nobody inherits anyone else's progress.

Somebody in the business has worked out how to get real value out of AI. They have found the prompt, the context, the skill, the way of asking AI that produces good work every time.

In single-player, that improvement is invisible to everyone else. It lives in their head, or it lives in their own folders and files. Context and intelligence is produced, but not shared.

Here is the test that illuminates whether a business is operating in single or multiplayer mode:

Documented intelligence and context is not the same as discoverable intelligence and context.

None of this is discoverable, so none one compounds across the business.

There is a subtler point to this and one that many confident AI users and businesses will ignore or misunderstand about the shared layer:

Shared context is not what two people already know, collected into one folder that they can both access.

Most of the context that changes decisions does not exist until people work the problem together.

The shared layer for a process is developed when it captures what the work actually produced, understood by the group, and is the starting point for anyone in the business when they next pick up that task. A drive full of uploaded documents is an input, it is not genuine context, or a well built context architecture. Because the context exists in the system for all to access, is not the same as the context being understood by all who can access it. That doesn't mean they should know the details of it all, but they do need to understand it exists, the way it can contribute to their world, and to be aware of how it might become an input into future work.

Let's bring that to life with an example we can all recognise and appreciate: the AI summary of a call.

There is an old line that covers this: the map is not the territory. The summary is a map of the conversation, and maps are useful, but nobody learns the terrain by reading one. The context a team actually runs on lives in the territory, in the working-through, where the nuggets and the gold actually sit.

A summary should start the conversation, not stand in for the detail.

Used the second way, it becomes one more single-player artefact that happens to be well written, fooling everyone into thinking it is all they need to know. And then we are into a matrix of summaries built on summaries built on summaries, each one a little further from the territory than the last.

That is the loss on one call. Now widen it out, because the same thing is happening to every scrap of hard-fought knowledge built up in the business.

Every business has a few people who have worked out how to get real value from AI. They have found the prompts, built the little tools, and their output is much better for it. Because none of that is inherited by anyone else, the person at the next desk is still starting from zero.

At its worst, single-player does something more damaging than waste that progress. It turns colleagues into rivals: people stop asking each other for help and start competing with each other's AI output, each presenting finished work the others had no part in, trusting the tool's answers over the person beside them.

Play that out across the whole business and four things follow.

The business does not compound in value. It just accumulates faster individuals.

The reason multiplayer is hard to picture is it's difficult to showcase. So here it is, in three ordinary workflows.

Start with sales proposals.

Take client discovery next.

Then content production.

It's the same move for each and every example. The tool is already there, but what changes is how the work gets done, it's redesigned so the context is shared across the business, informing the entire org, with a learning loop that closes behind it.

If multiplayer is so much more valuable, why do so few businesses reach it?

Not because they are lazy or under-trained. Because the crossing is hard for structural reasons, and the strategy that got them to single-player mode is the wrong strategy for what comes next.

The first mistake is optimising the worker instead of the work.

More training, more licences. It is the approach that visibly worked at single-player, so leaders do more of it. But training makes a person better at their own task. It does nothing to the shape of the work. Two hundred people can pass an AI course and the business runs exactly as before. This is the same mistake factories made in the 1890s. When electric motors arrived, most factories bolted a motor where the old steam engine used to sit and got a slightly faster version of the same factory. The transformation only came decades later, when someone rearranged the floor around what the motors made possible. The motor was never the point. The layout was, and redesigning the work around AI is where that begins (diagram I like to use in talks below).

The second mistake is leaving the context locked away.

Multiplayer runs on shared context, and most businesses have never treated context as something the business builds. They treat it as something people just have, in their heads and their own files, and so it never becomes the shared layer the work needs.

The third mistake is that nobody owns the crossing.

Single-player needs no owner, because each person runs their own. Multiplayer needs someone accountable for the redesigned workflow and context architecture. The default step is to hand it to the most enthusiastic AI person in the building, which is the same single-player instinct but in a different cloak. A keen individual cannot redesign how a whole team works. That takes a leadership mandate, not enthusiasm; AI expertise, not time on YouTube.

The fourth is the mandate itself: who is allowed to let AI act on the business's behalf.

The moment AI stops assisting a person and starts doing steps of the work, someone has to own that decision. Most businesses stall right here, and they stall because the organisation has not decided, not because the technology is not ready, mainly because people don't actually understand the work that needs to be done and what human-in-the-loop actually means.

Every one of these keeps the loop open. That is why the costs from earlier persist even in businesses that are doing plenty right.

Across Allexive's client work this year, training has bought about a fifth of the change. The rest has come from redesigning the work.

One caveat before making a sweeping change, because the argument is not that everything should be multiplayer.

The mode has to match the work, and one person's simple job often turns out to be another person's complex one, which is worth checking as a team before deciding we need to go multiplayer. The failure is running everything in single-player by default because nobody ever knew to ask the question, not the existence of single-player mode.

The strategy isn't to invest in more training and more licences. It is to pick one workflow, redesign it so the context is shared and the loop closes, so the next person inherits the last person's progress instead of starting cold.

That is the crossing in miniature, and it takes ongoing work, not a document written once.

Do that across enough workflows, by degrees, and the business changes character. It stops being a collection of faster individuals and starts being queryable, able to take what it learns in one place and use it everywhere.

That is the far end, and almost nobody is even aiming at it yet.

Not a tools problem. A work-design problem.

Not faster individuals. A business that inherits its own progress.

Not single player. Multiplayer.

Single-player makes a few people quicker. Multiplayer is the only mode where the whole business compounds.