AI coding assistants promise more code in less time. A recent telemetry analysis by Faros AI covering 22,000 developers and more than 4,000 teams shows where the time saved actually goes: under high AI usage, median pull request (PR) review time rose by 441.5 percent and the number of incidents per pull request rose by 242.7 percent (Faros AI, 2026). The bottleneck has shifted: from writing code to checking it.
For engineering leaders in medical technology, automotive, or financial services, this is especially relevant given strict regulatory requirements. When review capacity becomes the limiting factor, it hits exactly the role that is already scarcest on regulated projects: experienced engineers who take personal and professional responsibility for approvals.
AI democratizes the ability to produce code. It does not democratize, to the same degree, the ability to judge whether a change is correct within the overall system. A study of GitHub Copilot adoption in open-source projects by Xu and colleagues (2025) makes this difference concretely measurable: after adoption, experienced core developers reviewed 6.5 percent more code, while their own original code productivity dropped by 19 percent.
The additional output therefore does not land evenly across the team. It shifts toward the people who hold system knowledge, historical design decisions, and risk judgment. That is exactly what is hardest to delegate to an AI.
Developers themselves confirm this shift. In a longitudinal study of professional software developers, 82 percent reported spending less time on actual code writing by the second survey wave (Vella & Blincoe, 2026). The authors introduce a new category for this: “supervisory engineering work” – directing AI, evaluating its output, and correcting errors.
This changes what counts as valuable senior work. A principal engineer who types less code themselves can have become more productive, provided they deliver better specifications or build a verification rule that prevents hundreds of future defects. Classic metrics such as lines of code or raw PR count no longer capture this.
For promotion criteria, this is a blind spot in many organizations. Gartner predicts that by 2027, 70 percent of software engineering leadership job descriptions will explicitly require GenAI oversight, up from under 40 percent previously (Gartner, 2025). At the same time, Gartner explicitly warns against pushing junior talent out of shrinking team structures: doing so weakens knowledge transfer and the very experience base from which senior judgment later emerges (Gartner, 2026).
For your leadership team, this means concretely:
Organizations that fail to act on these points risk deploying their scarcest and most expensive expertise not toward architecture or knowledge transfer, but toward cleaning up a growing volume of AI output. Structured AI integration consulting and targeted training, such as the sepp.med Academy’s „ISTQB® Testing with Generative AI" course, address exactly this point, long before review backlogs turn into a quality risk.
How this shift affects not only team structures but also sourcing decisions and whether AI competence is perceived as an opportunity or a threat is more than a pure process question. This broader frame is exactly what the fourth talk slot at the next sepp.med Afterwork Exchange picks up: Markus Rautert, Chief Technology Officer at adidas, will speak in Röttenbach on September 24, 2026, about a gameplan for organizations navigating AI, from shifting delivery economics to changing talent pyramids.
We invite you to the next edition of the Afterwork Exchange. On September 24, 2026, decision-makers and experts will gather to discuss current trends in AI and software development. Join the discussion!
No, if anything the opposite. Gartner explicitly warns against hollowing out the talent pipeline if junior talent falls out of teams. Their experience-building shifts; it does not disappear.
Watch median review time, average PR size, and the share of pull requests merged without review. A noticeable increase over several sprints is an early warning sign.
Yes. Pure output metrics such as lines of code or PR count are losing explanatory power. Verification quality and architectural decisions are the more meaningful signals of seniority.
Tends to, yes, because approval responsibility there remains tied to a named individual and must be documented in an audit-ready way, regardless of how quickly the code was produced.
Firstname:
Lastname:
E-Mail Address:
Phone:
Subject:
Your message:
Yes, I consent to my personal data being collected and stored electronically. My data will only be used for the purpose of responding to my inquiry. I have taken note of the privacy policy.