The Workaround We Mistook for the Goal

All this time.

Time spent building systems and process and controls and governance all with the aim to elevate our capability and meet our increasingly varied needs. Those needs keep expanding, and each expansion demands more complex systems, process, controls and governance.

But we benefited. Faster, cheaper, wider. All driven by the information processing revolution. Everywhere around us data flowing in greater and more complex ecosystems, routed to smart orchestration capabilities providing real time capacity and demand, routed to automated robotic controls and increasingly AI capabilities. Everything smoothed over to provide information laden human workers with filters to absorb and process only what they needed to do their part of the work.

Yet it is still too much. We are too heavily laden and our brains constantly under pressure to absorb more, understand more and apply more, well beyond the capacity of the generations before us.

My question is simple. How much further can we expand and stretch ourselves. What is the breaking point of our human systems? How much longer can we hold the line and keep the pace?

I think we are approaching that breaking point blindly. We are not seeing but not understanding. The scale of change and acceleration we are experiencing is unsustainable.

We have to change.

Three standardisations, one constraint

I have been reading a pre-print called From Compressing Complexity to Accommodating Complexity: How AI Transforms Standardization and Individualization, and it discusses three standardisation's that happened over our recent history.

In systems terms, complexity is the sum of variation in inputs, variation in process and variation in outputs. I remember my dad saying to me you cannot be all things to all people and I think that applies to our historic systems as well.

The article covers three standardisations

Taylorism standardised labour. Break the work into defined steps, remove the variation between one worker and another, and you get predictable output from unpredictable people.

Fordism standardised the product. Any colour you like, provided the line only has to handle one.

ISO standardised the organisation itself, so that two companies who have never met can nonetheless rely on each other.

Different centuries, different domains, and underneath them one shared constraint. Processing difference was slow and expensive. Handling every input on its own terms was not affordable, so we removed the difference instead.

Each was a real achievement. None of them was the point. They were what you did when accommodating variety cost more than eliminating it.

The constraint is lifting

OK. So what, surely the standards we have in place are sufficient for our future?

Round about now, maybe, you are thinking, "those constraints were human and the standards and systems were put in place to help humans deal with complexity".

That was exactly my thought. So you may be questioning whether we still need all the standards now that Gen AI is here.

I believe we still need the standards and more. We need to have standards that promote and drive explicit human growth in an increasingly automated society.

Gen AI's ability to process complexity and variation is now at a stage that is outstripping human capacity and capability in many fields. Human processes depended on compressing complexity (losing depth in the process) to allow us to cope with the volume while Gen AI and related technologies are able to process huge volumes of information without compressing it.

The paper's phrase for this is the useful part: we are moving from compressing complexity to accommodating it. Compression was never neutral. Every standard threw away information about the individual case in exchange for something the system could actually handle. What is new is that we can now afford to keep some of what we were throwing away.

This is the point, and it is worth stating precisely, because standardisation was doing two jobs at once. One was a workaround for a capacity problem: we could not process variety, so we removed it. That job is ending. The other job was never about capacity at all, and it is the reason I want more standards rather than fewer. With AI extending our capability, the capacity workaround falls away and we can explore personalisation at a far deeper level.

I run a personal AI system I have named Jerry, built on Claude Code. It holds my writing rules, my working history and my standing instructions, and I use it across The Mantle Collective and most of what else I do.

The part worth reporting is not that its output is customised. It is what it declines to do. Jerry knows my rule that it must never invent my experience, so when this article needed an example of a standard that protects people, it did not write one. It left a marked gap and told me that paragraph needed something only I could supply. The HL7 example a few paragraphs down is what I put there.

A generic assistant would have produced a plausible healthcare anecdote and I might not have caught it. That is the difference personalisation actually makes: not a warmer tone, but a system that knows where its authority ends.

Why I would not sell the standards yet

Standards were never only a response to computational cost. They quietly carry four other things, and cheap computation supplies none of them.

Safety. A standard is a memory of how things went wrong. Clinical protocols, aviation checklists and building regulations encode incidents nobody wants to repeat. A system that adapts to the individual case can adapt itself straight past a hard-won boundary, and the adaptation will look like responsiveness right up until it does not.

Interoperability. Two organisations can work together because they agreed to be the same in a specific way. Personalisation is, by construction, the opposite of that agreement. If every system accommodates its own users perfectly and none of them share a shape, we have optimised each node and broken the network.

Accountability. You can audit a standard. You can ask whether the process was followed, and a yes or no means something. When the process is generated fresh for each case, "was this handled correctly" becomes a much harder question, and in a regulated setting it may become unanswerable.

Shared meaning. A standard is a common vocabulary. It lets people who have never met describe the same thing the same way. Lose it and you have not liberated anyone, you have made comparison impossible.

In practice. Healthcare systems still need a standard way to communicate with each other. This is HL7. The audit log of activity related to a patient protects the patient and the clinician. Two clinicians working across different domains speak a similar language.

It is the same for AI.

Accommodation is a choice, not a consequence

The other thing worth saying, is that none of this happens automatically. Cheap processing makes accommodating complexity possible. It does not make it happen.

Personalisation is not the same as being served. A system can hold enormous detail about you and use all of it against your interests. The technical capacity to accommodate variety is neutral. What we accommodate it for is not.

What I would advise

Stop treating your standards as a single category. Some of them exist because variation was expensive, and those are now open to challenge. Others exist because somebody got hurt, or because two systems have to speak, or because a regulator has to be able to check. Those are load-bearing, and cheap computation is not an argument against them.

Do that sorting before anyone proposes replacing templates with something adaptive, because the sorting is the actual work and it cannot be done by the system doing the adapting.

Then ask the harder question. If you no longer have to flatten people into an average, what would you build instead, and who would it be built for?

Sources

• From Compressing Complexity to Accommodating Complexity: How AI Transforms Standardization and Individualization, arxiv, preprint, not peer reviewed