Perly Consulting │ Beck Eco

The State of Play

A living index of AI adoption across industries — where established practice meets the bleeding edge
UPDATED DAILY

The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.

The Daily Dispatch

A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.

AI Maturity by Domain

Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail

DOMAIN
BLEEDING EDGEESTABLISHED

Formative feedback generation

LEADING EDGE

TRAJECTORY

Stalled

AI that provides detailed developmental feedback on student work, going beyond grades to guide improvement. Includes specific improvement suggestions and learning pathway recommendations; distinct from automated grading which scores rather than develops.

OVERVIEW

AI-generated formative feedback works well enough to deploy -- but not well enough to trust on its own. That tension defines the practice's leading-edge status. Forward-leaning districts and vendor platforms have moved from pilots to GA products, proving that LLMs can produce structured, actionable feedback on student work at a speed no human team can match. The value proposition is real: teachers reclaim hours, students get faster turnaround, and institutions can scale feedback across large cohorts. Yet the empirical record consistently shows that AI feedback remains inferior to human feedback on nuance, tone calibration, and adaptive support for struggling learners. Students, meanwhile, tend to overestimate AI feedback quality -- a source-credibility bias that compounds the accuracy problem. Production reliability adds another layer of risk; repeated model-drift and sycophancy incidents have forced rollbacks in deployed systems. The result is a practice that functions as a "teacher-amplifier" -- AI drafts feedback, humans validate it -- rather than an autonomous replacement. Most institutions have not yet adopted this approach, and those that have maintain mandatory human review. The question facing the field is no longer whether AI can generate feedback, but whether the quality and consistency gaps can close fast enough to justify the integration cost.

CURRENT LANDSCAPE

A growing cohort of vendor platforms and early-adopter institutions are operationalizing formative feedback systems at scale. Formative's Luna AI assistant, generally available since August 2025, has reached broad distribution across 90% of US school districts with 6+ billion student responses processed. Instructure's Canvas LMS released IgniteAI (April 2026), integrating rubric generation and feedback drafting into its core grading workflow -- evidence of ecosystem maturity as major LMS vendors embed formative feedback tools natively. Microsoft Teams Assignments ships AI Feedback Suggestions with explicit responsible-deployment guidelines. LearnWise reports 84% student preference for AI-generated feedback (40,000+ student sample) with deployment across Canvas, Moodle, Brightspace, and D2L. Wichita Public Schools (47,000+ students) and UK institutions piloting through the Jisc AI Assessment program demonstrate formative assessment as the primary deployed use case. These deployments maintain human review as mandatory workflow -- teachers review and edit all AI suggestions before students see them -- confirming the "teacher-amplifier" model as the operational standard, not an interim step.

The empirical picture remains stubbornly mixed despite operational scaling. April 2026 research confirms positive cognitive effects: a large-scale Frontiers study (n=1,079) shows AI precision feedback significantly enhances thinking ability (p<0.001) with intrinsic value identification mediating 32% of learning gains. Systematic reviews on L2 writing (55 studies) and automated feedback in HE (10 studies) identify collaborative tool use, custom design, and pedagogical scaffolding as critical success factors -- suggesting tool capability alone is insufficient without institutional redesign. Yet deployment quality remains contingent on assessment infrastructure. A 654-student peer-review study found half of participants flagged AI feedback inaccuracies, with only 6% preferring AI feedback alone. Experimental evidence from Chinese high schools showed AI-enabled visual feedback improving achievement but also increasing test anxiety. Reliability gaps persist across domains: Washington State University's study found ChatGPT accuracy on scientific hypotheses only ~60% (barely better than random chance) with 73% consistency across identical prompts. An MIT practitioner documented five spurious ChatGPT suggestions for every useful correction on feedback tasks.

The critical finding from March 2026 research: assessment design determines whether AI feedback drives learning or merely accelerates autopilot answer-completion. Qualitative evidence reveals that students with visible future accountability -- in-person exams requiring genuine understanding -- use AI feedback for reasoning and self-testing; those without accountability use it on autopilot. A 50-scholar synthesis identifies scalability benefits but forewarns of student dependency and quality consistency barriers; OECD research documents the performance-learning paradox: students write better essays with AI feedback but retain 80% less content, attributed to "fast AI" eliminating productive cognitive friction. A systematic review of 83 automated feedback studies confirms the field remains immature with heterogeneous results and inconsistent implementation. April 2026 Stanford research documents systematic demographic bias: high-achieving and White students receive developmental feedback while ELL/Hispanic students receive grammar-focused feedback, and low-achieving students experience feedback withholding. The trend line remains stalled: adoption has plateaued around the teacher-amplifier model, with quality consistency, equity gaps, assessment design contingency, and ROI realization blocking the path to broader uptake. The field consensus is clear: formative feedback generation succeeds only when embedded in pedagogically sound assessment systems with human oversight, not as a standalone tool.

TIER HISTORY

ResearchMar-2023 → Jul-2023
Bleeding EdgeJul-2023 → Oct-2025
Leading EdgeOct-2025 → present

EVIDENCE (87)

— William & Mary $300K GRI Accelerate grant-funded K-12 deployment of AI peer buddies that prompt reasoning and reflection rather than providing answers; focuses on critical thinking, equity, and teacher decision-making.

— Peer-reviewed white paper proposing theoretical reframing of sycophancy toward reflective responses; directly addresses feedback system design that acknowledges uncertainty and supports user autonomy.

— Meta-analysis of 72 studies showing AI teaching interventions yield significant positive effects (g_p=0.586) on effectiveness; clearly identifies boundary conditions and moderating factors enabling heterogeneous outcomes.

— Peer-reviewed research (Assessment & Evaluation in Higher Education, March 2026) with 10 principles for effective AI feedback in higher education; documents that students trust human feedback more and AI requires relational design.

— Real school district deploying adapted AI Assessment Scale framework across 20+ countries; teachers using framework to guide conversations about AI use, academic integrity, and demonstrations of learning.

— Stanford study (LAK best paper nominee, April 2026) documenting systematic demographic bias in AI writing feedback across 4 models; different tone and pedagogical expectations by student race, gender, achievement level.

— Expert analysis synthesizing research on feedback timing, specificity, and AI capability limits; emphasizes teacher judgment remains essential on creative and collaborative assessment despite AI routine assessment reliability.

AI Feedback & Grader | LearnWiseProduct Launches

— LMS-integrated product with 84% student preference for AI-generated rubric-aligned feedback; maintains teacher review and edit workflow; integrated across Canvas, Moodle, Brightspace, D2L demonstrating ecosystem maturity.

HISTORY

  • 2023-H1: Research and early-stage experimental deployment of AI-enabled formative feedback. Academic studies showed positive learning outcomes and student engagement with ChatGPT-based feedback; U.S. Department of Education released policy guidance on AI in formative assessment.
  • 2023-H2: Widening evidence base and platform integration. Multiple independent studies validated AI feedback effectiveness in international contexts (China, Estonia); Microsoft expanded Copilot with formative feedback capabilities to higher education. Research also documented significant limitations: GPT-3 feedback ineffective for struggling students, raising concerns about equity in deployment.
  • 2024-Q1: Transition toward operational commercial deployment. Studiosity's feedback service operating across Australian universities with measurable retention and GPA improvements; Microsoft Copilot for education launched with explicit formative feedback integration into Word and Teams; controlled research showed calibrated AI feedback effective but revealed critical need for human oversight. Academic critique emerged questioning trust and ethical implications, emphasizing that quality and equity remain unresolved challenges despite growing adoption.
  • 2024-Q2: Vendor acceleration met by empirical critique and institutional gatekeeping. Microsoft announced suggested AI feedback features in Copilot; Formative platform launched AI question generation. However, peer-reviewed research revealed human feedback superior across most quality dimensions, and educators documented reliability failures (grading inconsistency, bias). University of Sydney and other institutions implemented policies requiring human review and student transparency. Field consensus shifted toward AI as tool for educator review rather than autonomous feedback generation.
  • 2024-Q4: Mainstream adoption accelerates amid reliability and consistency concerns. AI classroom integration reaches 45-51% of educators and 86% of students with 93% of institutions planning expansion. Vendor platforms maintain core AI feedback features (Formative, Microsoft, Studiosity). However, empirical evidence reveals critical gaps: systematic review confirms ChatGPT inconsistency on subjective feedback; University of Pennsylvania study links ChatGPT feedback to 17% test score decline in high school math; production-level failure documented (ChatGPT model regression causing feedback application outage). Peer-reviewed research emphasizes need for teacher validation and oversight. Model reliability, fairness, and consistency emerge as blocking barriers to broader adoption.
  • 2025-Q1: Deployment research deepens and platform reliability becomes central concern. NSF-funded TeachFX research project launches with 300 teachers to test AI-enabled formative feedback for professional development; Microsoft announces Copilot Chat agents for tailored student coaching and feedback. However, empirical research exposes critical limitations: UC Irvine Nature Machine Intelligence study confirms systematic user miscalibration of LLM accuracy (overestimating reliability); domain-specific deployments show positive outcomes (English listening comprehension study, 60 learners) but require careful design. Industry debate shifts from "can AI generate feedback?" to "how reliable and equitable is it at scale?" Platform stability emerges as operational risk (ChatGPT service degradation documented February 2025). Field consensus: AI feedback tools require robust human oversight, transparent model limitations, and empirical validation before scaling, with reliability and fairness remaining gatekeeping barriers.
  • 2025-Q2: Empirical evidence accumulates on capability limits and quality gaps. Educator satisfaction studies confirm grading and essay feedback remain lowest-performing LLM capabilities (Copilot Chat essay grading rated 3.17/5). Peer-reviewed framework paper identifies systematic gaps in evaluation metrics and risks of overreliance. Small-scale empirical deployment shows positive outcomes (70-student Oman study) but requires context-specific design. Production reliability incidents continue (OpenAI GPT-4o sycophancy rollback May 2025), revealing alignment and feedback quality challenges. Deployment at scale remains cautious with institutional policies maintaining human review mandates. Field consensus solidifies: AI feedback as editor/reviewer tool for human educators, not autonomous feedback generator, with persistent barriers around consistency, calibration gaps, model drift, and bias requiring resolution before broader confidence.
  • 2025-Q3: Platform maturity and adoption acceleration meet evidence of persistent quality barriers. Formative releases Luna AI assistant for automated formative assessment (August 2025); teacher adoption reaches 60% using AI weekly with 6-hour time savings on grading. However, empirical studies from this quarter document critical concerns: doctoral and medical education studies reveal limitations in nuance, contextual understanding, and feedback tone (overly positive bias). OpenAI production incident (September 2025) rolling back ChatGPT update due to sycophantic feedback exposes alignment failures in deployed systems. Systematic review on AI in classroom assessment identifies unresolved equity gaps and bias risks limiting adoption in resource-constrained settings. Institutional gatekeeping continues with mandatory human review policies. Field consensus firms: AI feedback as tool for educator review, not autonomous generation, with feedback tone/honesty, reliability, and equity as blocking barriers.
  • 2025-Q4: Large-scale institutional deployment expands amid critical ROI and sustainability challenges. K-12 district rollout: Wichita Public Schools deploys Copilot for formative feedback and lesson planning across 47,000+ students with structured AI specialist guidance (December 2025). Microsoft Teams Assignments launches officially supported AI Feedback Suggestions with explicit responsible-deployment guidelines acknowledging model limitations (October 2025). Empirical evidence documents effectiveness gains: ChatGPT writing evaluation shows significant improvements in graduate-student writing mechanics and tone; qualitative research repositions AI as dialogic engagement partner. However, critical negative signals emerge: MIT analysis of 300+ enterprise AI deployments (December 2025) finds 95% deliver no measurable ROI due to workflow integration failures; Upwork reports AI agents fail 60-80% of standalone tasks. Preservice teacher research reveals affective barriers and concerns about feedback volume. Field consensus: AI feedback tools remain dependent on human educators for calibration, validation, and contextual judgment; widespread enterprise adoption challenges and slow ROI realization signal that formative feedback generation remains a "teacher-amplifier" practice with persistent reliability and integration barriers.
  • 2026-Jan: Empirical evidence clarifies capability limitations and institutional deployment patterns. Large-scale peer-reviewed study (n~500 STEM students) confirms AI feedback achieves comparable pedagogical quality to human feedback but reveals source-credibility bias—students overestimate AI feedback quality, undermining educator validation. MOOC research (n=161) documents positive student acceptance of ChatGPT-mediated feedback. Lab evaluation of 7 LLMs confirms feedback generation potential depends critically on rubric design and pedagogical scaffolding. Institutional research projects launch: Swiss AI Beacon project initiates design science R&D for AI-powered formative feedback system targeting 25,000 teachers and 300,000 students. Sussex University operationalizes AI as writing coach via custom GPTs. However, continued production reliability issues and ROI gaps persist, reinforcing consensus that formative feedback generation functions as teacher-amplifier (speed + first-draft generation) with persistent barriers around quality consistency, tone calibration, and cost-benefit at scale.
  • 2026-Feb: Adoption metrics consolidate and mixed empirical signals emerge. Formative platform reaches 90% of US school districts with 6+ billion student responses; Luna AI integration continues vendor-driven feature rollout. Experimental evidence from China confirms AI-enabled visual feedback improves achievement and self-efficacy over 13 weeks (125 high school students), but also increases test anxiety, indicating differential impact by learner profile. Large-scale peer study (n=654) on peer+AI feedback reveals continued preference for human feedback (58% prefer combined peer-AI, 36% peer alone) with 50% of students noting AI feedback inaccuracies. MIT practitioner reports systematic ChatGPT failures in feedback tasks (5 spurious suggestions per useful correction), reinforcing reliability concerns. Field consensus unchanged: AI feedback remains teacher-amplifier with persistent quality, reliability, and equity barriers limiting expansion beyond current penetration.
  • 2026-Mar: A 50-scholar multidisciplinary synthesis (CMU, Stanford, UC Berkeley) identifies scalability benefits for formative feedback but flags student dependency, quality consistency, and equity gaps as critical barriers. OECD research documents the performance-learning paradox—students write better essays with AI feedback but retain 80% less content, attributed to "fast AI" eliminating productive cognitive friction. Assessment design emerges as the decisive variable: qualitative research shows students with visible accountability use AI feedback for reasoning, while those without use it on autopilot. WSU peer-reviewed study finds ChatGPT accuracy on scientific hypotheses only ~60% (barely above random chance) with 73% consistency, underscoring fundamental reliability gaps. A systematic review of 83 automated feedback studies confirms the field remains immature with heterogeneous results; an RCT on developers shows AI-assisted groups scored 17% worse on comprehension despite identical task speed. Field consensus: effective formative feedback depends less on tool capability and more on surrounding pedagogical and assessment infrastructure.
  • 2026-Apr: Vendor platform maturity and empirical evidence on design-contingency consolidate the field. Instructure releases IgniteAI (April 10) with rubric generation, feedback drafting, and discussion insights—major LMS vendor GA signals ecosystem-level commitment. LearnWise reports 84% student preference for AI feedback with multi-LMS deployment. Jisc UK real-world pilot confirms formative assessment as primary use case with practitioners (London South Bank, Further Education colleges) reporting consistent, high-quality feedback and faster turnaround. Frontiers research (n=1,079) shows AI precision feedback significantly enhances cognitive development (p<0.001) with intrinsic value identification mediating effect. Systematic reviews of L2 writing (55 studies) and automated feedback in HE (10 studies) identify collaborative tool use, custom design, and metacognitive scaffolding as critical success factors. However, empirical concerns persist: Stanford study documents systematic demographic bias in AI feedback allocation; AIED 2026 study (1,349 instances, 117 teachers) finds 80% of AI feedback accepted without editing; Cambridge research documents AI EFL feedback covering only 8 of 16 human error types. Field consensus crystallizes around assessment-design contingency: AI feedback effectiveness depends less on tool capability and more on surrounding pedagogical infrastructure (rubric design, visible accountability, collaborative use patterns). Microsoft launches six new Copilot Teach features for formative tracking; $10M AmplifyGAIN research center (IES/NSF) begins RCT across 420+ teachers for scale evaluation (findings June 2027). High adoption (68% weekly use among K-12 teachers) coexists with persistent quality, equity, and consistency barriers confirming the teacher-amplifier model as the operational ceiling.
  • 2026-May: William and Mary received a $300K GRI Accelerate grant to deploy K-12 AI peer buddies that prompt reasoning rather than supply answers — a design direction explicitly responding to sycophancy and dependency concerns. A concurrent theoretical white paper formalises the sycophancy problem in feedback systems and proposes reflective, uncertainty-acknowledging responses as a design standard, while a meta-analysis of 72 AI teaching intervention studies (g_p=0.586) confirms positive average effects contingent on pedagogical boundary conditions.