The picture, above, is the console of an IBM System 370/168 — state of the art in computing in the mid-1970s. This particular one was recently retired. The operating system built for this particular machine continues to run workloads all over the world: from the MVS/370 of then to the z/OS of today the same batch jobs, CICS transactions, IMS workloads, etc. run as always. For those who are still invested in this paradigm, there seems little reason to change. It works — for them (for they are almost always organizations of a suitable size) it is cost-effective — and the cost and risk of replacing all that code is a thundering reason not to proceed.

Nevertheless, despite this opening this isn’t about the mainframe, or your other old systems. It is about the fact that the IT of the future won’t look anything like what we’ve built in the past. Meanwhile, the economic underpinnings of corporate IT are being steadily whittled away. If you work in IT in an enterprise, and you want to have a hand in the future, the time available to you is fast slipping away.

What The Future Looks Like, Technically: Essentially, application-less. To explain, begin with a metaphor: a notepad. It’s blank. To do something, you start writing on it. As you make notes, the updates you are making become messages on a message backbone. These are interpreted by a bevy of sensing routines. Some will initiate a search algorithm. Some will parse the data and offer new interpretations. Still others will try to match up elements and find supporting information, some of which is internal to your organization and some of which is external (for instance, writing “Mary Smith” and “604 892 0123” is more than enough to interpret the numbers as a telephone reference, look up a table of exchange assignments under the North American Numbering Plan, determine the city is “Squamish, BC” and then find that you have a “Mary Smith” in Squamish as a customer and make her record available for further matching. Note that we didn’t ever say what any of this data is: the routines in the “message handler soup” put forward interpretations. Other routines figured out which combinations were more or less likely to be fruitful.

This is how your brain works, more or less. It is a sense-making environment. We recognize a telephone number by the pattern of the digits: in the UK, for instance, if it (a) starts with a zero and (b) has four following digits, then that is a city code with a six-digit local number to follow — and the city code zeroes in on potential locations and post codes. In North America, the rhythm of three digits - three digits - four digits does the same job, with the first group isolating the area and the second group the probable location within that area (it is probable as numbers are more portable than they used to be). For a computer, it is just a string, and makes no “sense”, which is why we speak here of “sense-making” and “evaluation” routines, and a flurry of messages moving around.

“Oh, this will never work”, we’ve heard. We calmly ask in return if the speaker has ever used Google for a search. In essence, this is what is happening: hundreds of thousands of “search bits” dispatched to work on your request, other thousands evaluating and making sense of it all. In essence, there is no search database: there is a pool of internet sites captured earlier, coupled with massively parallel execution of relatively simple and hence well-tested routines.

Go a step further. Make each routine an object (so that it can keep a simple piece of data about itself). Have the object keep a counter of how many times “it” got selected as having the “right” outcome. Bias the evaluation routines to pay attention to the track record. Now you can have multiple different routines doing the same job — and a way to judge whether to trust the “orthodox” one (the one with the high score) or whether, this time, some “creativity” was needed (one with a low score).

This is grossly simplified, but it makes the point. There is no application here, yet a great deal of useful work can be done, since the process of transacting business, updating records, etc. is just more of what has been described.

Existing packages, applications, and so on don’t fit into this model. The goal here is to have an ecosystem that works in a massively parallel environment. No two “transactions” will likely follow the exact same path. No worries, though, because more routines can be used to compare this result to others, signal a possible issue, handle compliance, etc.

But it is a computing environment very, very different from what we know.

Today, we need to experiment with the core concepts. They would be ideal for a customer service station — something as simple as finding the customer’s records. In other words, our experiments can be incremental while we learn how to think differently about code.

The Financial Underpinnings Change, Too: If there’s one verity about IT financials, it is that they are going to be cut back. Today we need annual budget increases of 6-8% simply to stand still. To replace something big requires far more than that. We are immobilized, held back by everything that exists. The reason this is so is that changing the portfolio must be done in large, relatively indigestible chunks: there must be enough clear years to finance the project.

The new computing world doesn’t need that. The process of adding new routines is continuous, but each one is quite inexpensive by comparison. The “how did I do” process makes sure that unproductive innovations seldom affect operations. The massive parallelism in the environment allows for portions of the infrastructure to shut down and be replaced or restarted without cutting off the whole. It is therefore an environment that can be said to “learn”.

All of this fits into a corporate necessity: a far smaller share of SG&A going to IT. The existing computing model — whether done in house or outsourced — is just too expensive. (Given that the new model would act to absorb an acquisition — no real conversions required — it also will take over in any competition with the existing IT model. This will happen because it is a real competitive move.)

You may not believe, even now, that this is the future. Most people don’t. But we’re going there, and rapidly, despite any of our beliefs.