The use of evidence to inform teaching has become a priority in continuing professional development (CPD) and school improvement (Brown, 2015; Coldwell et al., 2017; Kennedy, 2014; Nutley et al., 2003; Owen et al., 2022; Slavin, 2020). Evidence-informed Practice (EIP) is based on the simple idea that instruction is strengthened when teachers draw on evidence of practice and relevant pedagogical research to guide what happens in their classrooms, rather than relying solely on their intuition or past experience (Brown, 2017; Nelson & Campbell, 2017). However, the enactment of EIP in schools remains uneven and difficult to sustain (Dagenais et al., 2012; Hemsley-Brown & Sharp, 2003; Malin & Brown, 2019).

Persistent barriers include issues of accessibility, relevance, and timeliness (Hering, 2016). Research is frequently inaccessible, locked behind paywalls or written in ways teachers find impractical (Schaik et al., 2018). Time constraints represent perhaps the most significant challenge (Cooper et al., 2017; Mandinach & Gummer, 2016). Meaningful engagement with classroom evidence and pedagogical research, and conducting authentic critical reflection is time-intensive and can exacerbate workload pressures in a profession already experiencing high levels of attrition and leadership turnover (Borman & Dowling, 2008; Finnigan et al., 2016). As a result, a persistent gap remains between researchers that produce scholarship and educators who ultimately decide what happens in classrooms, stunting professional learning and frustrating efforts to improve schools (Farley-Ripple et al., 2018).

This paper introduces Deliberative Instructional Agents (DIAs), a new class of educational technology for specialised instructional coaching that leverages advances in AI to make Evidence-informed Practice more feasible within the realities of everyday teaching. DIAs integrate the interrelated processes of evidence capture, research-informed analysis, and critical reflection into a single system. The result is an epistemic engine that efficiently situates teachers at the intersection of evidence of practice, pedagogical research, and their own professional experience, creating the conditions for situated professional learning as a matter of routine. In doing so, DIAs lay the groundwork for a new sociotechnical infrastructure for professional learning and school improvement (Lang et al., 2017).

To illustrate this potential, we present exploratory research using a DIA designed to support dialogic pedagogy, an instructional approach that emphasises learning through purposeful discussion, reasoning, and debate (Alexander, 2020; Barton, 2024; Knight, 2025; Mercer & Littleton, 2007). Specifically, the DIA integrates three innovative tools to support professional inquiry into dialogic teaching. Pierrot, a data collection tool, represents a technical breakthrough in speaker diarization, accurately transcribing and attributing speech during multi-person, in-person classroom discussions, without the need for specialised hardware. Cadence, the analytical component of the platform, automates transcript analysis using established research frameworks to distinguish between types of talk and applies an innovative mathematical model to identify critical instructional moments where there is opportunity for teachers to elicit deeper student engagement. Finally, Shape, an intelligent middleware, was developed to facilitate non-judgemental reflection on significant moments and trends in the learning analytics. This interactive function allows users to unpack the collated data and associated research through text-based conversation directly in the user dashboard, all while being guided by research-informed protocols for teacher noticing (van Es & Sherin, 2021; Mason, 2002).

How do teachers engage with evidence, analytics, and experiential knowledge during professional inquiry, and what does this reveal about how professional learning unfolds in practice? This paper will detail the DIA design and present results from exploratory research involving initial use of the proof-of-concept DIA. The research aims to advance theory by empirically examining professional learning as it unfolds through teachers’ engagement with evidence, analytics, and experiential knowledge during routine instructional inquiry.

Literature on teachers’ continuing professional development and learning is partial, fragmented, and under-theorised (Kennedy, 2014). Evidence-informed practice assumes that professional learning occurs when teachers engage with evidence of practice and relevant pedagogical research while also considering their own experiential knowledge. This study aims to develop a theory of professional learning by examining how a Deliberative Instructional Agent (DIA) captures teachers’ engagement with these different forms of evidence during structured reflection of teaching. The exploratory study involves maths teachers at two U.K. secondary schools (n=20). All participants attended workshops on dialogic pedagogy and received technical training to support classroom use of the DIA. Teachers then taught a series of three lessons at monthly intervals, each incorporating a dialogic instructional sequence. During each lesson, Pierrot ran in the background to generate a diarised transcript of classroom discourse. Following each lesson, Cadence provided teachers with a learning analytics dashboard highlighting opportunities to deploy discursive moves associated with Productive Disciplinary Engagement (PDE). PDE describes conditions under which students participate actively and meaningfully in disciplinary reasoning (Engle & Conant, 2002). In mathematics, dialogic teaching this is particularly relevant as teachers attempt to move beyond procedural instruction toward conceptual discussion. By foregrounding opportunities for PDE, Cadence intentionally draws teachers into productive struggle about their dialogic practice. The DIA then invites teachers to engage in text-based interaction with a highly specialised language model to critically reflect on these PDE opportunities. These interactions are captured as qualitative records of professional learning. Deductive coding is then used to identify episodes of teacher thinking preceding instructional decisions. Semi-structured interviews were subsequently conducted with all participants to supplement teacher inputs into the DIA, creating additional evidence of how teachers use data and knowledge to inform changes in practice.

Deliberative Instructional Agents (DIAs) represent a significant advance in the sociotechnical infrastructure supporting professional inquiry in schools. By capturing rich traces of classroom interaction and coupling these with AI-enabled analysis, DIAs materially alter what counts as useable evidence for teachers’ professional learning. Rather than rely on distal research findings or retrospective reflection, teachers gain timely access to evidence that is directly connected to their own instructional decisions and students’ disciplinary engagement. Crucially, this evidence is made available at a pace that aligns with teaching itself, allowing inquiry to be embedded within routine practice rather than positioned as an additional activity. In doing so, DIAs create new conditions under which evidence-informed practice can be sustained at scale. Beyond supporting individual decision making, DIAs also make professional learning more visible. Teachers’ engagement with learning analytics and their articulation of uncertainty through dialogic interaction with language models produce records of professional sensemaking. These records offer a novel form of evidence for understanding how professional inquiry unfolds during everyday practice, particularly during moments of productive struggle when instructional pathways are still open. This shifts the analytic focus from whether professional learning initiatives work to how teachers’ reason and learn as part of their everyday practice. This capacity is especially significant given the growing policy emphasis on structured forms of teacher inquiry, including lesson study, practitioner enquiry, and quality teaching rounds. While widely adopted, such approaches remain under-theorised, in part because the cognitive and interpretive work of teachers has been difficult to capture empirically. DIAs generate fine-grained evidence of professional learning processes, offering a powerful tool not only for school improvement, but for theory building in professional learning. This study demonstrates how this technology allows researchers to move beyond abstract models toward empirically grounded accounts of how teachers engage with evidence and experience uncertainty.

Alexander, R. (2020). A Dialogic Teaching Companion. Routledge. Barton, G. (2024). We need to talk. (p. 31). Oracy Education Commission. Brown, C. (2017). Achieving Evidence-Informed Policy and Practice in Education: EvidencED. Emerald Group Publishing. Cooper, A., Klinger, D. A., & McAdie, P. (2017). What do teachers need? An exploration of evidence-informed practice for classroom assessment in Ontario. Educational Research, 59(2), 190–208. https://doi.org/10.1080/00131881.2017.1310392 Engle, R. A., & Conant, F. R. (2002). Guiding Principles for Fostering Productive Disciplinary Engagement: Explaining an Emergent Argument in a Community of Learners Classroom. Cognition and Instruction, 20(4), 399–483. Farley-Ripple, E., May, H., Karpyn, A., Tilley, K., & McDonough, K. (2018). Rethinking Connections Between Research and Practice in Education: A Conceptual Framework. Educational Researcher, 47(4), 235–245. Hemsley-Brown, J., & Sharp, C. (2003). The Use of Research to Improve Professional Practice: A systematic review of the literature. Oxford Review of Education, 29(4), 449–471. Hering, J. G. (2016). Do we need “more research” or better implementation through knowledge brokering? Sustainability Science, 11(2), 363–369. Kennedy, A. (2014). Understanding continuing professional development: The need for theory to impact on policy and practice. Professional Development in Education, 40(5), 688–697. Knight, R. (2025). Classroom Talk: Evidence-based Teaching for Enquiring Teachers. Routledge. Lang, C., Siemens, G., Wise, A., & Gasevic, D. (2017). Handbook of Learning Analytics (First). Solar Research. Malderez, A. (2023). Mentoring Teachers: Supporting Learning, Wellbeing and Retention. Taylor & Francis. Malin, J. R., & Brown, C. (2019). Joining Worlds: Knowledge mobilization and evidence-informed practice. In The Role of Knowledge Brokers in Education: Connecting the Dots Between Research and Practice (pp. 1–12). Routledge. Mason, J. (2002). Researching your own practice: The discipline of noticing. Routledge. Mercer, N., & Littleton, K. (2007). Dialogue and the Development of Children’s Thinking: A Sociocultural Approach. Routledge. Nelson, J., & Campbell, C. (2017). Evidence-informed practice in education: Meanings and applications. Educational Research, 59(2), 127–135. Owen, K. L., Watkins, R. C., & Hughes, J. C. (2022). From evidence-informed to evidence-based: An evidence building framework for education. Review of Education, 10(1), e3342. Slavin, R. E. (2020). How evidence-based reform will transform research and practice in education. Educational Psychologist, 55(1), 21–31.