As AI takes on more of the work of writing code, Cinven’s portfolio CTOs are rethinking how software gets built, from the tools engineers use to the shape of the development lifecycle itself.

For software businesses, AI is no longer only a productivity tool for individual engineers. It is starting to change how the whole software development lifecycle (SDLC) is organised, from how work is specified to how it is built, tested and released.

Cinven works with technology leaders across its portfolio on this shift, including through a programme of AI forums for portfolio leadership groups. At a recent CTO forum, we were joined by Taxwell, Archer and other leading software businesses – they shared through how their teams are adopting AI, and the session featured in-house agentic development platforms in action. Their experiences, and the discussion that followed, point to several lessons for any business making the same transition.

Standardise a few things, early

AI coding assistants became widely available to businesses in 2023, and many portfolio companies began experimenting soon after. The step change came with the latest generation of frontier models and with portfolio companies began giving all engineers access and routing usage centrally so it could be secured and monitored.

Rather than mandate a single tool while the technology was moving so quickly, they standardised a small number of things. A common starting point was mandatory AI code review on every change, run in an adversarial mode with a strong emphasis on security. Engineers remain accountable for their code and its behaviour.

“The key to making real progress is to think big but deliver in small, concrete steps. A modest tool that engineers use every day and is gradually expanded is worth far more than an ambitious platform that never quite ships.” – Cristina Tuckley, AI & Tech Transformation Director, Cinven

As adoption grew, the focus moved from usage to impact. The heaviest AI users can take on markedly more ambitious work in a similar time, which raises a new question for the wider business: how product teams learn to scope bigger pieces of work.

Turn individual practice into a shared platform

The next step is to turn the best individual practice into a shared, repeatable process. Some have built proprietary platforms that take work through a structured pipeline of brainstorming, specification, planning, building, review and release, giving every developer access to the workflow of their most advanced AI users.

At the forum, one company showed this working live on a real feature request from one of its engineering directors. Its in-house AI agent identified the relevant part of the company’s software catalogue, asked the director scoping questions, drafted a specification and began building the feature directly into that product’s codebase.

Many see applying AI across their established systems, not just new ones, as the biggest opportunity. Incumbents already have the customers and brands that new entrants lack; the aim is to make sure their technology does not hold them back.

The lifecycle is being redrawn

Most organisations start with tooling, but the value depends as much on process and people. If engineers can write code several times faster, product and operations teams have to keep pace, and customers have to accept the change. A customer who saw their UI change multiple times a day may become frustrated.

Over time, we expect the lifecycle to converge into a continuous loop of intent, execution and monitoring, with humans setting outcomes and risk limits. The specification becomes the single source of truth that agents build and test against, pulling human effort towards the front of the process. However, for regulated businesses in particular, humans should stay in or on the loop rather than hand over full autonomy.

“As agents take on more of the build, the quality of what goes in matters more: clear specifications, well-documented systems and well-defined data. Testing and evaluation are critical to ensuring an agent’s output is fit to ship. The businesses that invest in those foundations now will see agent performance scale with each release.” – Stuart Walker, Partner, Head of AI & Tech Transformation, Cinven

Where the bottlenecks move

Once writing code is no longer the constraint, the pressure moves elsewhere: upstream, in deciding what is worth building; downstream, in releasing change at a pace customers can absorb. New features for users cannot always simply be A/B tested, and demand spikes narrow the window for experimentation.

“As AI removes the constraint on writing code, the harder questions are about what to build, how quickly customers can absorb change and how it is all governed. Those are leadership questions as much as technical ones.” – Tom Williams, AI & Tech Transformation Executive, Cinven

Looking ahead

The businesses that benefit most will be the ones that learn fastest: spotting what works in one team, making it standard across the organisation, and adjusting as the technology moves.

That is also why sharing experience matters. The CTO forum is part of a broader Cinven programme to share AI and technology practice across the portfolio, at a time when the technology is changing month by month.