For decades the constraint in software was always the same: never enough engineering capacity to build everything the business wanted. AI is dissolving that constraint. Coding assistants have grown into agents that can spec, write, test and deploy software, compressing work that once took weeks into days. Yet plenty of organisations have handed developers the new tools and seen very little change in results.

Faster code just moves the queue

AI-assisted coding is now table stakes. But when coding accelerates, the bottleneck shifts to whatever surrounds it — code review, planning cycles, security sign-off, and deciding what to build next. Speeding up one step of a pipeline doesn't shorten the pipeline; it moves the queue. A recent McKinsey discussion makes the point with an experiment: a streamlined squad of four, freed from the standard process, delivered in four days what a team four times the size took four weeks to complete. The difference wasn't the tool — it was the way of working.

What separates value from activity

• Redesign the whole lifecycle — from idea and requirements through testing and release, not just the coding step.
• Smaller, broader teams — generalists who can direct AI agents and own the output, rather than long chains of specialists.
• Standards built in up front — architecture and security embedded in the workflow, not inspected at the end.
• Humans on the “what” — when building becomes cheap, judgment about what is worth building becomes the scarce skill.

The pragmatic takeaway

Don't measure success by how fast developers type; ask where work waits. If your delivery process still looks like it did in 2023 — just with better autocomplete — the gains will stay on the screen and never reach the bottom line. The organisations capturing real value are redesigning teams, processes and risk posture around the technology, not simply installing it.