Today’s software teams are shipping faster than ever, but that speed comes at a cost.
According to a new Harvard Business Review Analytic Services survey sponsored by LaunchDarkly, over half of respondents say their organizations experience software release issues at least once a month.
And when things break, the impact can be massive: lost revenue, damaged reputation, burned-out teams. And AI is only raising the stakes.
The release risk nobody’s talking about
Most organizations think they’re managing release risk. But in many cases, their approaches are outdated. Manual rollbacks, big bang deploys, and reactive monitoring don’t just cause delays—they can actually be dangerous in today’s always-on, AI-driven market.
What’s worse is that only 6% of teams can detect release issues in real time. That means most organizations are flying blind, discovering problems after customers do.
Big bang is a big risk
The report highlights a trend that’s hard to ignore: companies are moving away from “big bang” deployments because bundling weeks of work into a single release magnifies risk. Bugs are harder to catch. Rollbacks are more painful. And if something fails, the area of impact can be huge.
The AI twist: more code, less control
Generative AI is now writing and powering code in production. That’s a massive leap—and a potentially massive liability. The survey reveals that half of the respondents believe AI increases software release risk. Model drift, unpredictable behavior, and increased release frequency are just a few of the new challenges teams are facing.
The shift: from reactive to resilient
Here’s what the best teams are doing differently:
- Decoupling deploy from release. They ship code without flipping it live. Feature flags are key here.
- Using progressive rollouts and targeting. They test in production, but with guardrails.
Phased, or progressive rollouts—especially when paired with real-time monitoring and targeting—are becoming the new normal. They let teams test changes with small audiences, catch issues early, and iterate quickly. It’s the kind of strategy that modern software teams (and their sleep schedules) depend on.
- Building observability into releases. They watch real-time metrics and catch anomalies before users do.
- Automating rollback. When something fails, recovery happens in seconds, not hours.