The Open Circuit
Suppose they're right. Suppose the general mind arrives, exactly as promised. The wall is still standing.
July 21, 2026
You can build a charge of any size.
Stack the voltage. A hundred volts, a thousand, a million. Hold it in the cloud, in the cell, in the capacitor. The potential is real, and it’s enormous.
It does no work.
Not until it finds a path to ground. Until the circuit closes, all that potential just sits there — charged, humming, useless. And here is the part that should stop you cold: raising the voltage does not help. A bigger charge across an open gap is not closer to doing work. It’s a bigger waste.
Hold that picture. It’s the whole argument.
The promise
The field is racing toward one thing, and it has been honest about it. Artificial general intelligence. A mind that reasons about anything. The general case, solved. Most of the money, most of the talent, most of the conviction in AI right now points at that one finish line.
I’m not going to argue about whether they’ll reach it.
Suppose they do. Suppose it arrives on schedule, exactly as advertised — a genuinely general intelligence, as capable as the most capable human across every domain there is.
Now watch what doesn’t change.
What generality can't touch
The model that can’t be trusted with one real file at your firm isn’t failing because it lacks generality. It’s failing because it doesn’t know your firm. Which clients are which. What your people mean by the words they reuse. Which of five records pointing at the same thing is the one the business actually treats as true.
None of that is a generality problem. It’s a particularity problem. And generality is the wrong axis to solve it on.
Make the mind as general as you like. That gap stays exactly as wide as it was. You raised the voltage. The circuit is still open.
This is the electrical fact underneath everything. Capability is voltage — potential to do work. Deployment is current — work actually done in the world. And current does not flow without a ground. The grounding layer is the path that closes the circuit between a general intelligence and a particular reality. No path, no flow. However high the charge.
AGI raises the charge to its theoretical ceiling. It does nothing about the path.
Why this makes the problem worse, not better
Here is the turn most people miss.
Right now, the grounding problem hides. When a system fails inside your business, there’s always an excuse ready: the model isn’t good enough yet. Wait for the next one. That line has absorbed the blame for years. It has let the field dodge the uncomfortable question.
AGI takes the excuse away.
When the model is, by definition, as capable as a model can be — and it still can’t be dropped into your institution and trusted — there is nowhere left to point but the gap. What remains when you’ve solved capability completely is the thing that was never about capability. AGI is the experiment that isolates the variable. It removes intelligence as a cause of failure, and what’s left standing, in full view, is grounding.
The smarter the model gets, the more obvious it becomes that smart was never the missing piece.
"But a general enough mind will ground itself"
This is the believers’ reply, and it deserves a straight answer.
The claim: sufficient intelligence dissolves the problem. A mind general enough will simply figure out your institution — infer your meanings, reconstruct your reality on its own. Grounding becomes automatic at the limit.
It doesn’t. And the reason is precise.
Your institution’s private meanings are not derivable. They are contingent facts about your specific reality — which name maps to which thing, which exception is load-bearing, which of two contradicting records the business honors. These aren’t truths a mind reasons its way to. They were decided, not discovered. And most of them were never written down.
A genius walking into your building does not know your building’s private language. Reads what’s posted faster than anyone alive. Cannot read what was never posted.
Intelligence accelerates the retrieval of what exists. It cannot retrieve what doesn’t — and in most institutions, the meaning doesn’t exist in retrievable form. It has to be built first. That was true at today’s capability. It stays true at infinite capability. The axis never intersects the problem.
What's actually left at the finish
So picture the line they’re sprinting toward. Now picture standing on it.
The charge is maxed. The general mind is real. And the circuit is still open — because all the money went into voltage and none of it went into the path. The work nobody wanted to do, the work that doesn’t scale and doesn’t demo, is exactly the work still sitting there undone. It didn’t shrink as the model grew.
It’s the residue. The part left over when capability is finished.
That’s the quiet bet hiding inside all of this. Not that AGI won’t arrive — maybe it will. But that arriving won’t feel like arriving, because the last problem left standing will be the one the whole race was built to avoid.
The charge was never the hard part.
The path to ground was.
This is the second and final piece in a two-part series on grounding — why the bottleneck in applied AI was never intelligence, but the layer that connects a general model to a particular world. Part one named the problem. This part follows it to the end of the capability curve, where it doesn't dissolve — it becomes the only thing left standing.
About the Author
Raghu Vishwanath is Managing Partner at Bluemind Solutions, a product engineering firm specializing in MRO master data governance. He writes about software engineering, AI, and building platforms that last.