Agentic Software Development
Leading engineering teams now treat AI agents as the primary producers of production code. The capability question is settled; the harder contest is whether organisations can govern, fund, secure and staff software that runs faster than the oversight built for it.
The evidence is no longer marginal. More than 80% of the production code Anthropic merges is now written by AI, Spotify merges over 650 agent pull requests a month, and the model frontier has advanced another full generation in a matter of months. Around this, leading teams have converged on spec-driven development: version-controlled specifications in place of ad-hoc prompts, codebases engineered to give agents the right context, and parallel agents backed by strong test suites and review agents as the safety net that makes the autonomy viable.
This paper argues that the pressure has shifted from the model to the organisation — cost, security, governance, procurement and people. It sets out where practice has genuinely converged, which hard questions remain open, and the deliberate moves needed to close the gap between machine-speed tools and human-speed organisations. The through-line is simple: speed amplifies both good design and bad decisions, so the teams that pull ahead will be the ones whose engineering and organisational fundamentals turn the acceleration into an advantage.
By Jayse Bergheim and Dr Rishni Ratnam, MXA Consulting — July 2026
Download the full article (PDF)
The evidence is no longer marginal. More than 80% of the production code Anthropic merges is now written by AI, Spotify merges over 650 agent pull requests a month, and the model frontier has advanced another full generation in a matter of months. Around this, leading teams have converged on spec-driven development: version-controlled specifications in place of ad-hoc prompts, codebases engineered to give agents the right context, and parallel agents backed by strong test suites and review agents as the safety net that makes the autonomy viable.
This paper argues that the pressure has shifted from the model to the organisation — cost, security, governance, procurement and people. It sets out where practice has genuinely converged, which hard questions remain open, and the deliberate moves needed to close the gap between machine-speed tools and human-speed organisations. The through-line is simple: speed amplifies both good design and bad decisions, so the teams that pull ahead will be the ones whose engineering and organisational fundamentals turn the acceleration into an advantage.
By Jayse Bergheim and Dr Rishni Ratnam, MXA Consulting — July 2026
Download the full article (PDF)
