AI has increased our ability to build software faster than our ability to understand what’s worth building. The competitive advantage long held by SaaS may shift from execution capacity to the quality of an organisation’s understanding, judgement, and conceptual models.
The next competitive advantage in software may not be the ability to build more, faster, but the ability to develop better models of reality and translate them into fundamentally better products.
Some thoughts.
Understanding is the bottleneck – great products emerge from deep domain knowledge, first-principles thinking, and imagining possibilities unconstrained by existing solutions. AI makes execution cheaper, but doesn’t automatically improve judgement. “Scratch your own itch” and “voice of customer” are direct relatives of this idea, and they rely on contact with reality.
Experimentation ≠ learning – one of the exciting applications of rapid prototyping is accelerated variation. We can build a dozen implementations and select for the one we like best. But experimental design is a known hard problem, and a poorly designed experiment can favour incremental improvement over idea maturation. If you go from idea to production before you have a chance to fall asleep that night, you might get trapped in local optima.
Abstraction creates leverage – great products compress underlying complexity into useful conceptual models. The breakthrough isn’t usually better functionality, but a fundamentally better way of “re-presenting” the problem. This links back to my first point on understanding.
High-falutin aside: Great art is sublime because it captures deep truth and conveys it back to us. This is the idea of resonance – when something outside of yourself vibrates in the same frequency as something deep within you, amplifying it.
Complexity is inevitable, but its distribution matters. The universe is only interesting because energy isn’t evenly distributed. The same is true of complexity – software should absorb without oversimplifying, where possible. And like biological evolution, cheaper software production will almost certainly lead to increasing specialisation and diversity of application (how many of your friends have their own workout app now?)
Legibility leads to blindness – organisations typically reward work where the impact is immediately obvious. This risks optimising the existing product while overlooking fundamentally better architectures or entirely different approaches. This again is a local optimisation problem.
Execution speed can deepen lock-in – Kodak became the most efficient producer of film and missed digital.
Organisations have different speeds of adaptation – each function operates on a different timescale determined by the value they product. Accelerating one layer doesn’t necessitate the same speed-up everywhere.
Implications
Optimise for learning not output – experiments should test and refine theories about customers and the world, not be in service of metrics.
Make conceptual architecture a leadership responsibility – explicitly model the domain itself, not just the existing offering. I think this is behind a huge amount of wasted effort in many businesses. The quality of your. conceptual model is upstream of your ability to effectively serve a market. Can everyone in your business recall, on demand, what you’re here to solve?
Go AI maximal on alternate futures. Don’t just concept your way through existing ideas, challenge the fundamentals and get weird. And on a similar note, apply scaled-up-AI-analysis to customer evidence – don’t just summarise a transcript, but hammer the underlying conceptual framework a customer has.
Be a coherence conservationist. Our ability to metabolise new concepts isn’t accelerating at the same rate as our software pipeline. Integrating change into a coherent strategy and model of the world is critical.
Generative Questions
Are we climbing the right mountain?
What is our fundamental model of reality?
What would make our product obselete?
Are we accumulating understanding or activity?
What becomes the bottleneck when building is nearly free?