“Perfection is finally attained not when there is no longer anything to add, but when there is no longer anything to take away.” — Antoine de Saint-Exupéry
AI gives us an unprecedented ability to add. The danger is that we begin to mistake accumulation for value.
Delivery is only the beginning (or be mindful of catabolic collapse)
Every new system and feature adds obligations: it must be operated, secured, monitored, documented, integrated, upgraded and eventually replaced or retired. Hackers love a juicy target, even if it’s that half-forgotten service that people are unsure whether it’s safe to turn off or not. If we respond to “cheaper” software creation by producing far more software, we may accumulate obligations faster than we acquire the capacity to discharge them. Under the weight of the proliferation of software, the organization starts to sacrifice its ability to build what it will need next to react effectively to changing market conditions and opportunities.
This is the dynamic described by catabolic collapse.
Catabolic collapse is a theory of societal decline in which a civilization accumulates more infrastructure than it can afford to maintain. Eventually, an increasing share of its available energy and resources is consumed merely preserving what already exists. Maintenance crowds out renewal. The society begins consuming its own capital simply to continue functioning. Think of debt payments taking up ever larger amounts of the national budget, the transport budget overwhelmed by the costs of fixing too many crumbling roads and bridges.
Lower the value threshold vs raise the ambition threshold
If we accept that every organization, even with AI, has a finite capacity to maintain software, then it follows that we should select carefully the software projects we commit to.
I can finally work on that feature that didn’t get funded time after time. I’m going to use AI to build it in two days rather than the estimated two weeks.
This is a case of lowering the value threshold and it’s a sloppy way to introduce one of the most transformational technologies in human history. You might get lucky this time, it might end up worthwhile, but then you equally might just be adding that extra bell or whistle, meanwhile your competitor is building a revolutionary new product that will blow you out of the water.
AI should raise the ambition threshold for software rather than lower the value threshold.
Tokenmaxxing (more tokens, more productivity!)
Unless you’re in a small, agile start-up, building a highly strategic product still requires a lot of cross-organizational work. Software engineers, researchers, product managers, market research and customer feedback, the list goes on. But forget all that, let’s reward our engineers (generally focused more on technology than business value) for using huge numbers of tokens to build stuff without careful evaluation of the actual ROI of the work.
It’s cool that Johnny finally rewrote that backend system in Rust, or rewrote the build system, or finally implemented that feature few customers actually are willing to pay for. But what was added may have done more for increasing the maintenance costs (and reducing the ability to react to future needs) than actually creating value.
Prototyping and demos are another slippery slope. Prototyping is an ideal case for AI with its ability to accelerate work. However, if the prototype represents a system that falls into the category of “previously too low-value to justify,” then the prototype is part of the same problem.
Raise the ambition threshold
It seems that in the initial euphoria at the turn of the year at seeing the new power at our fingertips, some conflated faster for cheaper, more for better. The lesson is that we should continue to apply sensible constraints to what we build. Just because we can build it doesn’t mean that we should.
The danger of using AI injudiciously is greater in large organizations, where the average worker is farther away from the customer and the business value. The more disconnected you are from the success and failure of the organization, the easier it is for tokenmaxxing to help you spend time and money on producing a lot of lower value work. Add the slopification of work and some organizations might actually see a net-negative impact. Indiscriminate token usage in the large enterprise is already showing signs of faltering as CTOs question the value of their AI usage mandates.
Business is a perpetual contest for advantage. Companies that spend their new AI capabilities trimming costs and burning down backlogs may soon be leapfrogged by competitors using them to attempt what was previously too difficult, risky or ambitious. So if you find that you are finally clearing all those nice-to-haves in the product backlog, ask yourself if your team is being ambitious enough.