Single-player AI and the multiplayer business

The work improves. That is the problem

There is rarely a single point at which a business identifies a single-player tool problem. It's much more subtle, more gradual and all the more difficult to see because of this.

An operations manager builds a prompt library that quotes jobs faster than the old process ever did. A finance assistant pastes an export into a free chatbot to get a summary before a board meeting. A marketer spends a year teaching a personal account the company's voice, and gets very good at it.

Each of these people is accomplishing the tasks they need in the best way they know possible. The tools are good and the work improves. Each team member is not looking to a future problem, or the business goals, they are just doing their job, in the here and now. That is precisely the problem. The failure here is structural, not personal, and nothing about it feels like a failure while it is happening.

In 2024, Microsoft and LinkedIn’s Work Trend Index found that among knowledge workers already using AI at work, 78% were bringing their own tools - Bring Your Own AI - rather than tools provided by their organisation ^S1. That is the denominator that matters: not “78% of all staff”, but most of the people already using AI at work.

The UK National Cyber Security Centre is clear about what follows when those tools are public consumer services: the query will be visible to the organisation providing the LLM, and the NCSC recommends not to include sensitive information in queries to public LLMs ^S2. And where personal data is involved, the ICO’s position is that UK data protection obligations still apply - accountability is not optional ^S3.

The trap - Do nothing and fall behind. Act badly and build a liability. Most businesses are told to choose, which is why most businesses stall.

Ban it, leave it, or take theirs

Ban the tools

A draconian option, but one that has some merit. The risk is real, and a policy is a cheaper control than an incident. What it fails to account for though is that the work does not stop, expectations on employee output are unchanged. It moves to personal devices and personal accounts, where the business loses the visibility it was trying to gain and the upside it never had. Worse, if competitors embrace AI tools effectively, you have now handicapped your business in the fight to lead the market.

Let it bloom

Adoption energy is precious and hard to manufacture and that is something that any smart business owner would want to protect. When employees want to innovate, want to deliver and improve the business, finding the way to support them is imperative.

What it fails to account for is that unmanaged adoption produces what we have labelled the five spikes - rogue builds no one approved, orphan systems whose builder has left, data leakage into personal accounts, one-way doors that cannot be undone, and key-person fault lines, where capability lives in one head.

In practice this is the path a capable person takes with a self-install agent framework such as Hermes or OpenClaw, and it works well right up to the point where a second, or a third person needs access to what it built.

None of these issues announce themselves. Each is discovered later, usually by someone who did not build it, and is responsible for delivering on a system that they likely do not understand and cannot recreate.

The AI already in the package

Embedded AI, typically from existing SaaS providers, delivers real scale, no setup, and genuine depth inside its own data. That strength though is its biggest weakness. Single-subject myopia presents a perspective of the business through one lens, its own module, and its memory, rules and roadmap belong to the vendor.

That is a trade-off that will be right for some businesses, not others and we explore further in our Comparison articles.

The business is the unit, not the login

Our design decision can be distilled to one sentence. Treat the business as the unit the system serves, not the person holding the login.

With that understanding more granular design decisions become self-evident.

Shared memory. One brain for the whole business, with every person and every agent seeing only their part of it. Context stops being something each individual accumulates privately and starts being something the company owns.

One gate. People and AI agents pass through the same permission system, and every action is written to a log that cannot be erased. An agent is not a special case with its own rules, it is another actor under the rules that already exist.

One goal. Every piece of work is checked against the Business Vision and North Star - the overall mission for the business or the relevant part of it, one active North Star per department. Alignment is a property of the architecture rather than a quarterly reminder, or a strategy day.

Heavier than signing in

Installing RedOS is heavier than installing an app, in so far as it doesn't simply creep quietly into the business. It wants delivery discipline and a Certified Consultant who knows the method, and it takes longer to stand up than signing in does. This is a good thing, but it is a trade-off.

The US Navy SEALs are credited with the expression "slow is smooth, smooth is fast" which at its core means that rushing leads to mistakes and wasted time, whereas deliberate controlled actions allow you to complete tasks safely and efficiently. RedOS is smooth!

For a genuine solo operator it is the wrong choice, single-player AI is the right tool for a single player, and not to say so would simply be dishonest.

RedOS is our implementation of those three ideas, delivered as a sovereign install. For the right business, we back it as the right choice, but that doesn't mean it's for everyone.

When someone leaves, the memory stays

When someone leaves, the memory stays. The prompts, the context and the judgement encoded in them belong to the business, so a resignation is a staffing event rather than a capability loss.

When an auditor, an insurer or a buyer asks who can see what, and what the AI has done, the answer is a log rather than a shrug. Diligence stops being an archaeology exercise.

When agents multiply, and they will, they pull in one direction, one that the business owner has chosen, because the goal is in the architecture rather than in the quickly forgotten briefing.

None of this requires better tools than the ones already on the market. It requires the business, rather than the individual, to be the thing the system is built around.