Part 1 of 2. The second post introduces Argus and what to do about all this. Read Part 2 →

Superintelligence has moved from the tech press to the evening news. If you run a business you are hearing about it from friends, from your board, and from your kids, usually wrapped in either panic or a sales pitch. I want to explain what is actually going on in plain terms, what it is not, and why it matters to someone whose job is to keep customers happy and the company solvent.

The reason people think it is coming faster than expected has a name that sounds like science fiction and is quietly becoming an engineering fact. Recursive self-improvement, or RSI, is the idea that an AI system helps build a better AI system, which then helps build a better one still. Each generation arrives faster and cheaper than the one before, because more of the work of building it was done by the previous generation.

For a long time this was a thought experiment. This year the leading labs started saying out loud that it is partly happening. Their models write most of the code. They produce a lot of their own training data. They debug their own training runs and find improvements that get fed back into the pipeline. People still decide what to research and what to release, but the fraction of the work that is automated goes up every quarter.

Nobody serious claims the loop is closed. The piece still missing is judgment about which problems are worth solving. What the labs do say is that the direction is set, and the disagreement is about how many years it takes rather than whether it happens.

Intelligence, action, and accountability

So what changes for a company like yours if that keeps going?

The first thing is that intelligence stops being scarce. Right now, hiring smart people is hard and expensive and slow. Within a few years most of the knowledge work in a business will be doable by AI agents you rent by the hour or the task. Drafting proposals, following up leads, chasing invoices, managing a delivery schedule, and preparing a renewal all become things an agent can run from start to finish, and every one of your competitors will have access to the same agents.

The second thing is that the number of actions taken in your name explodes. Today your team might touch a few hundred customer interactions a week. A handful of agents can take thousands of actions a day. Any one of them can promise something, upset someone, or create a liability. The agent will not be accountable for that. You will.

The third thing is that the world's institutions will not keep up. Regulators, auditors, boards, and your own customers will all start asking the same question, which is how you know what your AI is doing and whether you can stand behind it. Companies with a good answer will be allowed to move fast. Companies without one will be slowed down by their own risk people, or will move fast and get hurt.

Beyond doom and hype

That is the real story, and it is neither the doom version nor the hype version.

The doom version says superintelligence makes your business irrelevant. It does not. A model that is smarter than every employee you have still does not know that your biggest client said on a call last Tuesday that they are unhappy with the onboarding, or that Maya at the fund asked about new deals last month. It knows the world in general. It does not know your company.

The hype version says you just need to plug in the latest model and everything sorts itself out. It does not. An agent working from your CRM alone has no idea what was agreed in the meeting. An agent reading one inbox does not know about the decision taken on the call. Agents act on the evidence they can reach, and in most companies the evidence is scattered across a dozen places and mostly out of reach.

What competitors cannot rent

A few months ago a client asked me the question that sits underneath both versions. If in a few years anyone can rent agents that are better than his best people at most of what those people do, what does his company have left that a competitor cannot also rent?

When every company rents the same frontier, the intelligence itself stops being a differentiator. Everyone's proposals will be well written. Everyone's outreach will be timely. Everyone's analysis will be sharp. What will separate companies is whether anyone can trust what their agents are doing, and whether the company can show its work when asked.

Consider what happens to a customer relationship when agents run it. One agent handles the outreach. Another manages the delivery schedule. A third prepares the renewal. Each of them works from whatever system it was pointed at, so each has a partial picture. Then the customer calls and asks why they were told two different things about the launch date, or why the renewal quote ignores the delay everyone knew about. A person has to answer that call, because customers do not accept apologies from software.

Three gaps to close

I think of the problems that open up here as three gaps.

The first is a truth gap. No single system knows what was promised, what was delivered, what changed, and what the customer currently believes. At human speed that is an annoyance you work around with a phone call. At agent speed it produces conflicting commitments and duplicated outreach faster than anyone can catch them.

The second is a supervision gap. No manager can review thousands of agent actions a day. Without somewhere that ranks those actions by how much they matter, flags the odd ones, and routes the high-stakes ones to a person before they happen, leadership is stuck between trusting everything and slowing everything down, and both of those lose.

The third is a learning gap. A frontier model does not know which of your accounts tend to churn after a delivery slip, or which contact at the client actually decides on the renewal, or what a healthy relationship looks like in your market. That knowledge comes from your own outcomes over time, and it is the one thing your competitor cannot buy from the same lab.

What your company keeps

So, back to my client's question. He has his evidence, meaning the record of everything his company has said to and heard from its customers, connected and understood. He has his judgment about those relationships, built up from his own outcomes rather than the internet's. And he has the ability to look a customer or a regulator in the eye and say exactly what was done in his name, by whom or by what, and why.

None of that comes from a model. It comes from owning the layer above the model, and the awkward truth is that most companies do not own it yet, because their evidence is scattered across calls, inboxes, meetings, and people's heads.

That is the real implication of superintelligence for a business. Control, supervision, and context become the things that matter most, and most companies have none of the three in a form they could point to. The second post is about what that looks like on the ground, and about what we built at Soligence to fix it.

Next: Your company already knows the answer. It just cannot find it.