Parameter count is the new horsepower

Nobody has ever selected accounting software on calculations per second. Nobody compares customer systems by database size. In every other category of business software we worked out long ago that the specification sheet is a poor proxy for whether the thing solves your problem, and we ask about workflows instead.

Then AI arrived and we started comparing parameter counts like teenagers comparing biceps.

The number does mean something, to researchers, in context, alongside a dozen other measurements. What it does not do is tell you whether the thing will handle your document types, integrate with your systems, or cost less than the process it replaces. It is horsepower, and horsepower has never told anyone whether a car can get down a narrow lane without taking the wing mirrors off.

The release cadence is the actual damage#

The specification argument is silly but harmless. The release schedule is what genuinely costs organisations money, and it does so without anyone deciding anything.

Consider a finance director who signs off capital investment without visible anxiety. Put an AI decision in front of the same person and they freeze, because between starting the evaluation and reaching a recommendation, three newer options have launched, each announcement carrying the implication that yesterday's choice was already wrong.

So the evaluation restarts. Then it restarts again. Eighteen months later there is a thick folder, no deployment, and a growing suspicion that waiting is the responsible thing to do, when in fact waiting is simply the thing that requires no decision.

Nobody engineered this paralysis deliberately, but it is not commercially inconvenient for the people producing the announcements. Perpetual anticipation is an excellent way to keep a market attentive.

The organisations hurt most are not the ones with money to experiment. It is the mid-sized business that has been told AI is now essential to survival, is trying to act responsibly, and finds that the ground moves faster than their procurement process can complete a lap.

The boring work is where the money is#

While all this is going on, actual value is sitting in places nobody writes announcements about.

Invoice handling. Document classification. Extracting figures from forms that arrive in eleven formats from four hundred suppliers. Routing correspondence to the right team. Checking one system against another and flagging the differences. None of it exciting. All of it repetitive, expensive in human hours, and well within the reach of technology that has been stable for a while.

A well-implemented document process can take a meaningful bite out of processing time and pay for itself within months. It will not generate a press release. It will show up in the operations budget as a number that is smaller than last year, and the people who used to do the tedious part will be doing something else.

That work does not need the newest model. In most cases it barely needs the second newest.

Stability is a feature nobody prices#

There is something the current pace ignores completely, and it will look obvious in retrospect.

Stable systems can be tested properly. They can be documented, and the documentation stays true. They can be integrated, and the integration keeps working. Staff can be trained without their knowledge expiring on a schedule set by somebody else's marketing calendar. Auditors can be shown how something works and the explanation will still hold at the next audit.

Every serious enterprise technology went through this. Databases did not reinvent themselves fortnightly. Business systems did not ship a new architecture every Thursday. They became boring, and becoming boring was the thing that made them buildable-upon, because you cannot lay foundations on something that keeps changing shape underneath you.

We will get there with this too. We are simply in the loud part, and the loud part always feels permanent while you are inside it.

The calculator test#

There is one comparison I keep coming back to because it clarifies the standard we have quietly accepted.

Imagine a calculator that gave the wrong answer two per cent of the time, improved from three per cent last year. Would you celebrate the improvement? You would not. You would bin it, and you would think slightly less of anyone who had recommended it, because a tool that is confidently wrong occasionally is worse than no tool, since you will act on it.

That is roughly the standard being applauded in some deployments of this technology. Not because anyone is dishonest, but because the comparison being made is against last year's version rather than against the requirement. Progress against yourself is not the same as fitness for purpose, and only one of those matters when it is inside your accounts payable.

What to do on Monday#

Ignore the releases and start from your own problems. Which processes consume disproportionate human time? Where do errors reliably creep in? What does somebody redo every week? Those questions are stable, they will still be the right questions next year, and the answers determine which technology you need rather than the reverse.

Then demand demonstrated value rather than described potential. Any supplier pushing the newest thing should be able to explain precisely why it solves your specific problem better than the established option. If they cannot make that connection concretely, they are selling research and you are funding it.

And value stability and support over novelty, because your operations team will live with this for years after the excitement passes and whoever sold it has moved on.

The market will mature. The announcements will slow to a sensible rhythm, the valuations will find a level, and this period will be remembered slightly sheepishly.

Until then the objective has not changed and never was winning anything. It is better operations, lower costs, and customers who are dealt with properly. That is almost always achieved with proven, stable, faintly boring technology that simply works, and there is no shame at all in getting to the shops in something sensible.