The Computing Community Consortium’s (CCC) latest report, Beyond Code: Engineering Trustworthy Software Systems with AI at Scale, takes up the question of what has to change in software engineering once implementation is largely automated. With adoption of AI coding tools now reported at 84% to 90% of developers, much of the work shifts to specifying, verifying, and maintaining what AI produces. This report, however, documents a widening gap between what AI-assisted development is expected to deliver and what these systems can currently do — as well as between the sheer amount of code being generated and the current ability to verify it. The report presents a way forward for closing these gaps.

Beyond Code: Engineering Trustworthy Software Systems with AI at Scale was authored by Randal Burns (Johns Hopkins University), Sebastian Elbaum (University of Virginia), Gabrielle Allen (University of Wyoming), Nils Aschenbruck (Osnabrück University), Terry Benzel (University of Southern California), William Gropp (University of Illinois Urbana-Champaign), Rick Kazman (University of Hawaii), and Manish Parashar (University of Utah).

Read the Report Here

Critical Paths Forward for Researchers

The report identifies seven technical directions for the computing research community:

• Bridge the intent gap by developing models and workflows capable of interactive requirement elicitation, formalized specifications that capture quality attributes, and traceability mechanisms that record why design choices were made
• Pursue neuro-symbolic integration, pairing generative models with symbolic reasoning so that code and its formal proofs are produced together and verified automatically
• Develop cross-layer orchestration through higher-level languages designed for generative agents, formal interfaces for inter-agent communication, and IDEs that function as orchestration dashboards
• Make security intrinsic to generation through goal-driven security interfaces, secure-by-default models, and cross-AI validation that uses one model to probe another for vulnerabilities
• Establish training data transparency and IP governance, including membership inference, provenance tracking, and models architected for verifiable attribution
• Advance transpilation as a robustness strategy, forcing generated output into safe and verifiable representations and providing equivalence guarantees across hardware platforms
• Extend AI to deployment, developing agents that reason about package contents and dependencies, monitor federated environments, and remediate anomalies without manual intervention

Shifts Specific to Academia

Alongside these technical directions, the report recommends changes to how academic research in this area positions itself:

• Move industry partnerships past benchmark creation
  • Current industrial-academic alliances largely position academia as a supplier of evaluation benchmarks, limiting research impact and constraining the talent pipeline. A deeper partnership would give researchers the production context, telemetry, and scale that building and testing next-generation models and tooling now requires, and give industry a partner working on problems that sit behind corporate walls.
• Aim at the next generation of models rather than evaluating the current one
  • The report suggests building models from the ground up for formal verification, security-by-construction, and long-term maintainability, which is work that corporate labs routinely bypass under market pressure. University labs retain a lead in safety-related work, including the sociotechnical and ethical effects of AI, which gives them an advantage.
• Rebuild curricula around requirements, specifications, and system design
  • Because students will no longer build mental models through the act of programming, new pedagogical approaches are needed that cultivate system-level intuition and complexity management. Computational thinking itself also needs redefinition to include managing higher-level abstractions and anticipating the behavior of agentic systems.

Read the Full Report

The findings of this report are informed by discussions at the CCC Beyond Code: Engineering Trustworthy Software Systems with AI at Scale visioning workshop held February 25-26, 2026 in San Francisco, CA, generously supported by the National Science Foundation (NSF) and IEEE Computer Society. The workshop convened 41 experts across academia, industry, and government in artificial intelligence, software engineering, programming languages, cybersecurity, and systems engineering. 

We encourage all members of the computing research community to read the full report at the link below.

Read Beyond Code: Engineering Trustworthy Software Systems with AI at Scale

Tune in to the CCC LinkedIn Showcase Page for updates and more reports like this. Stay connected with CCC for the latest insights, publications, and opportunities to engage by subscribing here.

This material is based upon work supported by the U.S. National Science Foundation (NSF) under Award Nos. 2300842 and 2619366. These awards support the Computing Community Consortium (CCC), a programmatic committee of the Computing Research Association (CRA). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.