Daniel Russo, Ph.D.

A deep curiosity, channelled into research that matters to academics and to practice.

Featured directions
Flagship

AI in Software Engineering

AI is changing software development. Whether it changes it in the ways the conversation assumes is an empirical question. This line of work produced a consistent finding: workflow compatibility predicts adoption more than perceived usefulness.

AdoptionCreativityEthics
Active programme AI4SE1DK

Human-Centred AI for Software Engineering

30.4MDKK budget
14partners
10K+practitioners
2025–29duration
Grand Solutions
Innovation Fund Denmark
FIG 1.1 HACAF structural model (Russo, 2024, ACM TOSEM) Path diagram of the structural model of the Human-AI Collaboration and Adaptation Framework (HACAF), estimated with PLS-SEM on 183 software engineers. Perceptions about the Technology predicts Compatibility Factors (0.766, p < 0.001), and Compatibility Factors predicts Intention to Use (0.536, p < 0.001); this is the highlighted path. Perceptions about the Technology also predicts Social Factors (0.405, p < 0.001), and Personal and Environmental Factors predicts Perceptions about the Technology (0.313, p < 0.001). The direct paths to Intention to Use from Perceptions about the Technology (0.155), Social Factors (0.087) and Personal and Environmental Factors (minus 0.004) are not significant. Variance explained (R squared): Compatibility Factors 0.587, Intention to Use 0.489, Social Factors 0.164, Perceptions about the Technology 0.098. 0.313*** −0.004 n.s. 0.155 n.s. 0.766*** 0.536*** 0.405*** 0.087 n.s. Personal and Environmental Factors Perceptions about the Technology R² = 0.098 Intention to Use R² = 0.489 Compatibility Factors R² = 0.587 Social Factors R² = 0.164 FIG 1.1 HACAF structural model. PLS-SEM, standardised path coefficients with R² for endogenous constructs; bootstrapping with 5,000 subsamples; n = 183 software engineers. *** p < 0.001; n.s. p > 0.05, dashed. Highlighted: the only significant direct predictor of Intention to Use. Source: Russo (2024), ACM TOSEM 33(5), Fig. 2, Tables 13, 14. doi:10.1145/3652154. CC BY 4.0, redrawn.
FIG 1.1 HACAF structural model, core paths (Russo, 2024, ACM TOSEM) Path diagram of the core paths of the Human-AI Collaboration and Adaptation Framework (HACAF), estimated with PLS-SEM on 183 software engineers. Perceptions about the Technology predicts Compatibility Factors (0.766, p < 0.001), and Compatibility Factors predicts Intention to Use (0.536, p < 0.001); this is the highlighted path. The direct path from Perceptions about the Technology to Intention to Use (0.155) is not significant. Variance explained (R squared): Compatibility Factors 0.587, Intention to Use 0.489. Four further paths of the full model are listed as text below the diagram. 0.155 n.s. 0.766*** 0.536*** Perceptions about the Technology (PT) Compatibility Factors (CF) R² = 0.587 Intention to Use (IU) R² = 0.489 Also in the full model, not drawn here: PEF PT 0.313*** PT SF 0.405*** PEF IU −0.004 n.s. SF IU 0.087 n.s. PEF: Personal and Environmental Factors. SF: Social Factors. R² from the full model. FIG 1.1 HACAF structural model, core paths. PLS-SEM, standardised path coefficients; n = 183 software engineers. *** p < 0.001; n.s. p > 0.05, dashed. Highlighted: the only significant direct predictor of Intention to Use. Source: Russo (2024), ACM TOSEM 33(5), Fig. 2, Tables 13, 14. doi:10.1145/3652154. CC BY 4.0, redrawn.
Emerging direction

AI-Native Software Ecosystems

When autonomous coding agents contribute to shared codebases at scale, the ecosystem behaves in ways no single agent explains. This line treats AI-native software as a complex adaptive system and works toward an emergence science for multi-agent coordination: theory first, then measurement. First evidence, from more than 930,000 agent-authored pull requests: integration risk is a property of the repository, not the agent.

EmergenceMulti-Agent CoordinationEcosystem Governance
Preprints 2026 AI-Native Ecosystems

An emergence science for multi-agent coordination

930K+agent-authored PRs
2×friction concentration vs humans
~50%variance at repository level
7falsifiable propositions
arXiv 2604.19827, 2606.28235
The three pillars
01

AI in Software Engineering

What the evidence shows about AI in engineering teams: adoption dynamics, disruptive technology studied rigorously, and principles for responsible integration.
02

People, Teams & Organisations

Across agile transformation, hybrid work, and team diversity, interpersonal and organisational variables predict outcomes more reliably than technical choices.
03

Education & the Next Engineer

From computational thinking in schools to cooperative learning at university, active and participatory formats outperform passive ones at every level.
Foundations

The Methodological Base

PLS-SEM, Empirical Standards, guidelines for LLM-assisted research, and evaluation frameworks that pair validated self-report instruments with objective telemetry: the analytical infrastructure that makes the findings in each pillar trustworthy.

Open by default

Every study is open-access; data and replication packages are deposited in Zenodo. Profiles: Google Scholar, ORCID.

There is no end. There is no beginning. There is only the infinite passion of life.
Federico Fellini

Empirical research is built with partners.

Every finding on this page comes from studying real teams in real organisations. Funders and partners have included the Alfred P. Sloan Foundation, the Carlsberg Foundation, Innovation Fund Denmark, defence agencies, and multinational banks. In a typical collaboration, the partner contributes field access, data, or funding; I contribute research design, validated instruments, and analysis; the findings are published open access. A seven-year collaboration with The Liberators shows the shape this can take.