Software program engineering restricted by lack of full automation

A survey by Coleman Parkes for Harness has discovered that the usage of synthetic intelligence (AI) in software program improvement is on the rise, however there are gaps, which means that extra AI and automation could possibly be deployed.

The survey outcomes, revealed in Harness’s State of AI in software program engineering 2025 report, present that almost all of software program improvement groups imagine software program supply will likely be dominated by AI brokers working alongside human engineers inside 5 years.

Nearly two-thirds of respondents stated they’re utilizing AI for code technology, 60% have used AI for documentation, and 57% are utilizing AI for high quality assurance and testing. Different areas of software program improvement presently supported by AI have been discovered to incorporate error remediation (55%), safety compliance (54%), and efficiency and value optimisation (53%).

Areas the place the software program builders polled are seeing enhancements from utilizing AI embrace the pace of code creation (51%), sooner testing and high quality assurance (45%), and developer onboarding time (43%).

The survey discovered that, on common, organisations use eight to 10 distinct AI instruments for software program improvement. Some are utilizing far bigger numbers of instruments, which suggests there’s a threat of AI device sprawl, introducing complexity that would lengthen the time it takes to get new members of the software program improvement engineering staff absolutely onboard.

Harness famous that device sprawl and vibe coding can amplify operational threat. It warned that fragmented toolchains and inexperienced builders utilising AI assistants are creating governance challenges, rising incidents and incurring hidden prices. It really helpful that IT leaders consolidate instruments right into a unified platform and set up AI-powered guardrails to scale back complexity and hold groups centered on innovation.

In accordance with the Harness research, organisations have excessive AI use for coding, however immature testing, deployment and governance. Harness really helpful that IT leaders pair AI coding assistants with automated testing, deployment verification and safety checks to forestall threat, price overruns and guide toil.

The ballot means that automation maturity is the primary barrier, limiting the pace with which software program engineering groups can ship software program. The largest efficiency hole is just not in code creation, however in supply. In accordance with the Harness survey, steady supply (CD) and governance stay under-automated.

The pace increase from AI-assisted coding is making a stress wave that’s crashing towards a wall of under-automated, legacy downstream processes. Whereas builders are writing code sooner than ever, the techniques meant to check, safe and deploy that code are struggling to maintain up. Solely 6% of the IT professionals surveyed stated their organisation’s CD processes are absolutely automated. This has led to the emergence of what Harness calls an AI velocity paradox.

For organisations with lower than 1 / 4 of their CD workflows automated, solely 26% have seen a rise within the frequency at which code is shipped to manufacturing from their use of AI coding instruments. This jumps to 57% in those who have between one and three-quarters of CD processes automated. In accordance with Harness, transferring from low to reasonable automation in CD, subsequently, greater than doubles the probability that organisations will see a velocity acquire from AI coding instruments.

Harness recommends that IT leaders spend money on downstream automation to translate AI-driven code velocity into enterprise velocity.