This started as a fairly practical observation.
Over the past few months, we have worked on a number of operational problems in non-software businesses: fragmented data, difficult reconciliation processes, manual workflows, and decisions that depended on pulling information together from systems that were never really designed to work together.
The problems were different, but they tended to share one characteristic: a few years ago, most would not have been obvious candidates for bespoke software. There was rarely a good economic case for building something specifically for them.
So businesses worked around them. Sometimes quite effectively. Sometimes for years.
What is changing is the economics of that trade-off.
AI-assisted software engineering is starting to shift that equation. Over the past year, LLMs and coding agents have become genuinely useful across much more of the software engineering lifecycle. Not just writing code, but investigating existing systems, testing, documentation, infrastructure, debugging and support.
That does not remove the need for experienced engineers. The difficult parts remain: understanding the operating problem, deciding what should actually be built, navigating messy systems and data, and knowing where not to over-engineer.
What has changed is the leverage of that judgement. A small, experienced team can investigate and resolve problems with a different equation of time, team size and cost.
And that changes something more important than development speed: it changes the range of problems that are worth solving with software.
There is a large category of operational issues in non-software businesses that sit in an awkward middle ground. Important enough to matter. Too specific for packaged software. Not large enough to justify a traditional technology programme.
For private equity investors, this changes the role software engineering can play.
It can increasingly be treated as a capability that is deployed selectively against operational problems, rather than as something that sits only inside technology companies or large IT functions.
We have seen this firsthand. In one infrastructure business, a small engineering team built a tool in roughly three months to consolidate fragmented electricity-grid data and support network-capacity, site-selection and investment decisions.
In a FibreCo, a similarly small team addressed billing and reconciliation issues across contracts, network data and billing records in a matter of weeks.
Neither required a new enterprise platform. They required people who could understand the problem, work through the systems and data underneath it, and build the smallest useful solution.
For investors, that potentially turns software engineering into a repeatable value-creation capability: one that can be deployed across individual businesses and, over time, across a portfolio.
That is the part of AI-assisted software engineering I find most interesting.
Not that software becomes faster to build.
That more problems become worth building software for.