Software Insights · Evidence-Based Newsletter
Research You Can Act On. Every Week.
Software Insights translates peer-reviewed software engineering research into actionable guidance for engineering leaders, CTOs, and practitioners who want to make better, evidence-informed decisions.
Software Insights is the only newsletter that sits at the intersection of peer-reviewed software engineering research and practitioner-accessible writing. Each issue takes a recent empirical study and unpacks what it means for teams shipping software today. No abstract-speak. No opinion dressed as evidence. Just rigorous findings rendered in clear language, with concrete recommendations you can apply in your next sprint, hiring cycle, or strategy review.
The newsletter publishes weekly on Steady. Below is the complete archive. Subscribe to receive new issues directly.
Published Issues
Your Agents Pass Every Test. Your System Can Still Fail.
Organisations building multi-agent AI pipelines face a measurement problem that standard benchmarking cannot resolve: individual agents can pass every capability test in your suite while the system they compose fails in ways that no single-agent analysis would predict. This issue unpacks the Gradient Institute’s taxonomy of multi-agent failure modes and what it means for your governance frameworks.
Read full issueCutting the Junior Pipeline Before the Automation Arrives
Harvard researchers analysed 284,974 U.S. firms and found that junior employment at GenAI-adopting companies fell roughly 9 percent in under two years. The mechanism is not displacement but pre-emptive hiring freezes. This issue examines what that means for knowledge transfer, senior cognitive load, and the talent architecture that organisations will need when the anticipated automation actually arrives.
Read full issueAI Can Solve Your Benchmark. Can It Fix Your Codebase?
A new study from METR proposes the 50%-task-completion time horizon and applies it across twelve frontier models released between 2019 and 2025. The doubling time is approximately 207 days. But the benchmark baseline measures a low-context human working isolated tasks from scratch, not the senior engineer who built the system. This issue unpacks the 5–18x context gap and what it means for tooling and staffing decisions.
Read full issue56% of Organizations Get Nothing from AI Because Compatibility Was an Afterthought
Data from PwC, MIT, Deloitte, and McKinsey converge on the same finding: the majority of organisations see zero measurable returns from their AI deployments. Only 12% achieve simultaneous revenue gains and cost reductions. This issue introduces a theoretical frame drawn from organisational compatibility research that explains the pattern and identifies the structural conditions that separate the 12% from the rest.
Read full issueThe New Engineering Judgment: Five Skills That Survive the Agentic Transition
A software researcher at University College London built a production-grade, multi-module data analysis platform in five days without writing a single line of code. The prompt-level empirical record reveals which human capabilities remain indispensable: system decomposition, failure diagnosis, quality judgment, architectural foresight, and integration testing. This issue examines each through the lens of the empirical data.
Read full issueWhen Code Becomes the Output: Rethinking What We Teach (and Hire For)
A synthesis of industry roadmaps and complexity science research suggests software engineering education is teaching the right skills for the wrong paradigm. As AI generates increasing volumes of code, the bottleneck shifts from writing to evaluating, integrating, and governing. This issue examines what that means for curricula, hiring criteria, and the competency frameworks that underpin engineering career ladders.
Read full issueSpeed at the Cost of Quality: The Hidden Tax of Agentic AI
Agentic AI tools promise dramatic speed gains in software development, but emerging research reveals a consistent pattern: velocity improvements often come at the expense of code quality, maintainability, and long-term system health. This issue examines the empirical evidence on the speed-quality trade-off and identifies the conditions under which agentic AI delivers genuine productivity rather than deferred technical debt.
Read full issueThe “Country of Geniuses” Requires a Constitution, Not Just a Login
The “country of geniuses” arrives 2027: 50M agents, Nobel-level IQ, 100x speed. They will game your tests. You need a constitution, not just a login. This issue examines what governance frameworks organisations need to build before autonomous AI agents become embedded in production engineering workflows and decision-making processes.
Read full issueDavos 2026: From “Chat” to “Work”
Davos 2026 data shows 87% of enterprises adopted AI, but only 31% see measurable ROI. The problem is not the model; it is the fit. This issue synthesises the key findings from the World Economic Forum’s annual meeting and maps them to the empirical literature on technology adoption, organisational readiness, and the gap between executive optimism and engineering reality.
Read full issueEurope’s Tech Renaissance Beyond the “Kill Switch” Narrative
Three years ago, most “cloud strategy” discussions in Europe sounded the same: Where are the best engineers? Which vendor gives the biggest discount? Can we move fast enough? This issue examines how European technology sovereignty is evolving beyond regulatory compliance toward genuine strategic capability in AI infrastructure, talent development, and industrial policy.
Read full issueTechnological Sovereignty in a New Era: A Compass for Europe’s Digital Future
2026 marks the end of the “Strategic Promissory Note” era of enterprise AI. This inaugural issue examines the convergence of European digital sovereignty policy, AI governance frameworks, and the practical implications for engineering organisations operating across jurisdictions. A foundational evidence brief for leaders navigating the regulatory and strategic landscape of European technology independence.
Read full issueSoftware Insights
Research That Changes How You Build Software
Every week, one peer-reviewed study. Translated into clear language. With concrete guidance you can apply on Monday. Subscribe to Software Insights and join engineering leaders who make evidence-informed decisions.