AI belongs inside the project record

Actor: Construction Dive / Bain

Construction Dive interviewed Bain partner Peter Hanbury about the capital required to build AI infrastructure, citing a possible $5 trillion to $6.5 trillion of data-center spending through 2030 and about $1.5 trillion in annual infrastructure spending by 2031.

The interview identifies power constraints, electrical labor, long material lead times and community opposition as immediate construction limits; Bain said at least 75 projects worth $130 billion were blocked or delayed in the first quarter of 2026.

These are Bain projections and reported market conditions, not a productivity result from an AI construction tool. The construction-specific AI signal is that the buildout depends on decisions about power, procurement, labor and acceptance long before a facility is operating.

Why it matters: AI demand is creating a construction pipeline whose bottlenecks are physical, regulatory and social rather than purely computational. That changes what owners and contractors must measure before counting backlog.

Practical AI use case or operational implication: Create a pursuit register that links power availability, long-lead equipment, electrical labor, permits, community conditions and commissioning dates to each AI-facility milestone.

Suggested executive takeaway: Treat the $6 trillion figure as a scenario, not committed work; price the constraints and delay exposure that determine which projects can actually proceed.

Actor: Western Sydney University, Monash University and collaborators

A 62-page systematic review analyzed 50 journal articles published from 2021 through 2026 on agentic AI in construction, including reasoning, planning, memory, tool use, digital twins, robotics and Internet of Things systems.

The review finds applications across planning, scheduling, logistics, safety, inspection, quality control, structural design and facility operation, but says most studies remain at prototype or simulation stage with inconsistent evaluation and limited real-world deployment.

The authors propose a Human-Governed Construction Agentic Intelligence Loop that keeps context, perception, planning, tool choice, human-governed action, evaluation and learning connected. It is a framework, not evidence that autonomous project delivery is ready.

Why it matters: The most useful finding is not that agents exist; it is that construction needs a repeatable control loop around them. Prototype volume without field validation should not be mistaken for delivery maturity.

Practical AI use case or operational implication: Use the H-CAIL logic to define a pilot: input records, allowed tools, decision owner, escalation threshold, evaluation metric, audit record and learning boundary.

Suggested executive takeaway: Use research maturity as a gate. Require project-level validation, interoperability and qualified oversight before an agent can alter a schedule, design, safety record or commercial commitment.

Actor: Togal.AI

Togal.AI launched Construction's Last Exam, a benchmark covering 33 construction tasks such as reading plans, measuring spaces and materials, and producing bid and takeoff numbers.

The release describes tasks in which a model measures every balcony in a plan and is scored against correct measurements. Togal says it built construction-specific vision models from tens of millions of plans rather than relying only on general language models.

The benchmark is announced by the vendor and its performance results are not supplied in the source. Its value is the evaluation design: blueprint understanding and data handling need construction-specific tests rather than generic model scores.

Why it matters: Estimating and design teams need a common way to compare systems on the mistakes that affect quantities and bids. A construction benchmark moves procurement away from fluent demonstrations.

Practical AI use case or operational implication: Run two candidate tools on a closed plan set, preserve the ground-truth quantities, classify errors by drawing type and measure review time before considering production use.

Suggested executive takeaway: Ask every drawing-AI vendor for task-level accuracy, data controls, reproducibility and failure examples; do not use a benchmark announcement as proof of takeoff reliability.

Actor: TRUERESTORE / TRUEBUILT Software

TRUERESTORE said 130 contractor customers have created 50,000 projects and that Ripple, its AI project coordinator, is now included in every job on the platform.

Ripple works from moisture readings, drying logs, LiDAR scans, 360 photos, walkthrough videos and crew notes. It can answer job questions, flag missing documentation, draft a scope, update status and generate reports, with users confirming changes first.

The announcement is a vendor account, not an independent estimate of claims acceptance, cycle-time or payment improvement. Its construction-specific evidence is the structured field record and the confirmation boundary around AI-generated work.

Why it matters: The event shows that an agent becomes useful when it knows the project record and can identify missing evidence while crews are still on site. The data foundation is the differentiator, not the chat interface.

Practical AI use case or operational implication: Pilot one restoration job type by comparing Ripple's missing-evidence flags and draft scopes with a coordinator's review, then record corrections and carrier feedback.

Suggested executive takeaway: Keep source readings and human confirmation authoritative; measure documentation completeness and time to estimate before expanding agent permissions.

Actor: vBrief / AEC Magazine

AEC Magazine reports that vBrief uses AI to help construction teams recover valuable information from inboxes, folders and disconnected project records.

The platform's proposition is to give project information a more usable structure so teams can find the context behind decisions rather than treating files as isolated documents. The article presents the product direction, not a named project outcome.

This is a product account and does not establish extraction accuracy, permission behavior or a reduction in RFIs. The construction test is whether the system preserves document version, responsible party and decision context.

Why it matters: Search is not the same as project intelligence. An AI information layer matters only if its answer points back to the governing record and does not erase the uncertainty around it.

Practical AI use case or operational implication: Use a closed project archive to test retrieval of one RFI, change order or submittal, checking source version, access rights and reviewer corrections.

Suggested executive takeaway: Require citations, version awareness, exportability and permission tests before using an AI information layer in a contractual or safety workflow.

Actor: Vectorworks / AEC Magazine

AEC Magazine describes Vectorworks 2027 as a release aimed at practical design workflows, with enhanced performance, visualization and expanded support for Revit interoperability.

For construction teams, the relevant issue is whether improved exchange and model performance preserve design intent as information moves between authoring tools, coordination platforms and downstream production.

The article does not establish an AI deployment, field productivity result or error-rate improvement. It is a construction technology event whose AI relevance is the data and workflow context future assistants will depend on.

Why it matters: AI cannot rescue a design process that loses objects, revisions or constraints at file boundaries. Interoperability remains a prerequisite for any reliable construction automation.

Practical AI use case or operational implication: Test one live exchange between Vectorworks, Revit and the coordination environment, comparing geometry, classifications, references and revision history.

Suggested executive takeaway: Treat interoperability claims as a measurable acceptance criterion and keep the native authoring record authoritative until a project-level test passes.

Actor: OpenAI / AEC Magazine

AEC Magazine reviews experiments in which OpenAI's GPT-6 Astra turns drawings, photographs and instructions into 3D geometry through Blender, Rhino and Revit connections.

The article reports impressive first models but also failures involving circulation, DirectShape output, framing, furring and component choices. A BIM Pure test produced a recognisable cabin while leaving a model with limited native editing behavior.

The evidence is experimental and includes vendor or developer demonstrations without independent verification. The article explicitly says the results do not establish dependable judgment about clearances, circulation or construction.

Why it matters: A fast first model is not a coordinated building model. The planning value is in exploring options while keeping the design decisions, model semantics and corrections visible.

Practical AI use case or operational implication: Use a non-issued option study with a read-only connector first, then compare native elements, constraints and revision behavior against a human baseline.

Suggested executive takeaway: Require professional review, native-model fidelity, change persistence and licensing review before generated geometry enters a coordinated or issued model.

Actor: Formas.AI / AEC Magazine

AEC Magazine reports that Formas.AI launched Cartesian as a native agentic 3D modeller designed to produce editable geometry for physical-world applications.

The proposition is different from image generation: an agent works with geometry and design intent in a modelling environment. For AEC, that could support early optioning, but downstream constraints, object semantics and construction documentation still govern usefulness.

The launch is a product claim, not a project benchmark. No independent evidence in the source proves code compliance, constructability, quantity fidelity or safe automated issuance.

Why it matters: Agentic modelling is most useful upstream, where teams can explore alternatives without confusing a compelling shape with an approved building solution.

Practical AI use case or operational implication: Run one bounded massing or layout study and compare editable geometry, dimensions, assumptions and export behavior with the office's normal process.

Suggested executive takeaway: Keep Cartesian in feasibility until interoperability, constraint handling and review evidence support a production design decision.

Actor: Pirros / AEC Magazine

Pirros added an AI-powered Quality Management System that can review an entire Revit project against codes, firm standards, project specifications and model performance.

Findings can be assigned to people or to Mira, Pirros' AI agent, which can make a proposed fix in the model. The system surfaces findings in context, does not auto-apply them and keeps an audit trail of who resolved what and when.

The source describes capability and an early beta user, not independent precision, recall or rework reduction. The controlled-application boundary is stronger than an unattended model edit, but it still needs project validation.

Why it matters: QA/QC becomes more actionable when a finding stays attached to the model element and resolution record rather than becoming a disconnected spreadsheet item.

Practical AI use case or operational implication: Load one representative Revit package, compare AI findings with the firm's checklist, and measure false positives, correction time and audit completeness.

Suggested executive takeaway: Require code-owner review, versioned standards and non-automatic application until model-level accuracy is demonstrated on the firm's project types.

Actor: Volve / AEC Magazine

Volve added an orchestration layer to its construction tendering and preconstruction platform so users can ask it to compare information across specifications, drawings and an IFC model.

The system is described as determining the steps needed for a check, running comparisons and returning findings with their source. That is aimed at the document and model contradictions that estimators must resolve before pricing.

The article does not report independent takeoff accuracy, missed discrepancies or bid outcomes. Source-linked findings still require an estimator to inspect revision, scope and governing document status.

Why it matters: Preconstruction AI has value when it surfaces a material disagreement before the firm carries it into a bid, not when it merely summarizes a tender package.

Practical AI use case or operational implication: Run one trade package in parallel with the current estimate and classify every finding as a true issue, source mismatch or false positive.

Suggested executive takeaway: Keep the bid basis human-controlled and require source citations, revision handling and documented false-positive performance before award decisions.

Actor: Buildcheck / AEC Magazine

Buildcheck secured $12 million in Series A funding for a vision-based drawing-intelligence platform intended to catch issues, understand revisions and apply contractor QA/QC playbooks before construction.

The platform checks drawing health, 2D clashes, omissions, logical rules and spec-to-drawing contradictions from 2D PDFs without requiring a BIM model. The company says in-house experts review outputs before customers receive them.

The source cites customers and product capabilities but no independent performance study. Human review is a useful control, yet buyers still need to test severity, missed issues and revision behavior on their own drawings.

Why it matters: Drawing review is an estimating control because an omitted or contradictory condition becomes a quantity, scope and risk assumption. Vision-based review can widen coverage if its findings remain inspectable.

Practical AI use case or operational implication: Use a closed drawing set with known issues, compare Buildcheck findings with the estimator's review and log the cost or scope effect of each accepted flag.

Suggested executive takeaway: Procure against documented issue classes, reviewer qualifications, data retention and correction rates rather than the funding headline.

Actor: Kora Studio / AEC Magazine

AEC Magazine reports that Kora Studio 1.0 is a Revit plug-in for generative façade design and is promoted as enabling rapid exploration of complete façade options.

A faster option study can affect early quantities, panelization, material choices and constructability questions, but only if the generated alternatives retain dimensions, openings, performance assumptions and the inputs used to create them.

The launch does not establish a project-level cost, carbon or schedule result. It is an early-design aid, not evidence that a generated façade is ready for pricing or engineering release.

Why it matters: Estimating improves when design alternatives arrive with enough structure to compare them, not when the team receives attractive images detached from quantities and constraints.

Practical AI use case or operational implication: Generate a small façade option set and compare panel counts, openings, materials, performance assumptions and manual cleanup against the current concept process.

Suggested executive takeaway: Use Kora for bounded optioning until the firm can trace each priced quantity to a reviewed, revision-controlled design.

Actor: BIMlogiq / AEC Magazine

AEC Magazine reports that BIMlogiq's Argus is a multi-agent platform intended to automate Revit workflows from family creation through rebar and model production tasks.

The platform's project-controls relevance is the coordination of repeatable model operations and the possibility of assigning specialized tasks to agents rather than asking one assistant to perform an opaque sequence.

The launch does not provide a schedule baseline, error study or production project outcome. Multi-agent orchestration adds a need to record task order, inputs, permissions and recovery behavior.

Why it matters: An agent chain is a project-control system only when the team can see which step changed the model and can stop or reverse a bad handoff.

Practical AI use case or operational implication: Test one repetitive model task with versioned copies, record agent actions and compare total human review time with the manual baseline.

Suggested executive takeaway: Require task-level logs, rollback and a named model owner before connecting agent chains to a live project schedule.

Actor: Cupix / AEC Magazine

CupixWorks 5.0 is described as linking reality data, project information and asset records into an Enterprise Spatial Intelligence layer that extends from construction into operations.

For project controls, the important handoff is a spatial record that can connect what was captured on site with the issue, asset, milestone or handover information that explains what should exist there.

The article describes a product direction rather than an independent schedule or closeout result. Teams still need to validate capture currency, location accuracy, data ownership and the status of accepted information.

Why it matters: Reality data becomes a control input when it is current enough to change tomorrow's work and structured enough to survive the handover to operations.

Practical AI use case or operational implication: Pilot one workfront or asset package, reconcile dated captures to the schedule and issue log, and measure variance-correction time.

Suggested executive takeaway: Keep the approved baseline and acceptance record authoritative; use spatial intelligence to expose variance, not to silently rewrite it.

Actor: OpenSpace / AEC Magazine

AEC Magazine reports that OpenSpace is extending its reality-capture platform with AI agents intended to see, understand and act on jobsite information.

The control-room implication is a move from storing images to answering project questions against visual evidence, potentially helping teams identify progress, missing work or issues earlier in the day.

This is a platform announcement and does not establish independent progress accuracy, schedule recovery or safe autonomous actions. Any agent must preserve capture date, location, source image and reviewer disposition.

Why it matters: The practical value of visual AI is not a more searchable photo library; it is a timely, reviewable variance signal connected to the project control process.

Practical AI use case or operational implication: Run an agent in shadow mode on one area, compare its findings with superintendent records and measure correction, escalation and action latency.

Suggested executive takeaway: Keep human acceptance over schedule changes and issue closure until project-level false positives and missed conditions are known.

Actor: All3

All3 unveiled a second-generation Mantis quadruped designed to carry up to 100 kilograms, climb stairs with a 45-kilogram arm and locate and drive structural screws without an operator in the loop for that task.

The company reports 8-of-8 screw placement in repeated demonstrations to 0.5 millimetre tool-tip precision, plus a deterministic 1,000-times-per-second control loop and a controlled response to power loss on stairs.

These are All3 tests, and the safety assessment is still in progress. The first live construction deployment is planned for early 2027, so no production-site reliability or incident result is established.

Why it matters: Robotics is crossing from capture to installation, but the site—not the controlled demonstration—will determine whether force, navigation, recovery and supervision are adequate.

Practical AI use case or operational implication: Choose one repetitive fastening or handling task, define exclusion zones and recovery procedures, and compare intervention, cycle time, quality and near misses with a manual baseline.

Suggested executive takeaway: Demand a task-level safety case, operator-rescue plan and evidence from comparable active sites before treating Mantis as production capacity.

Actor: Structured AI / AEC Magazine

Structured AI now lets teams review IFC or native Revit models alongside PDFs, specifications, codes, manuals and schedules, with AI agents checking those sources against each other.

A finding can be cited to an exact sheet or model location, opened beside the affected Revit element and proposed for application only after an engineer approves it. The platform advertises more than 900 checks across disciplines.

The article reports capability but no independent precision, recall or field rework reduction. Its operational value depends on whether cited findings are complete, correctly located and resolved against the governing revision.

Why it matters: Field teams benefit when a quality issue can be handed from the drawing to the model and then to a controlled correction without recreating the context manually.

Practical AI use case or operational implication: Run one coordination package in a non-issued environment, compare findings with the field and design team's accepted list, and record missed and nuisance issues.

Suggested executive takeaway: Require engineer approval, immutable source references and a clear distinction between a proposed fix and an accepted change.

Actor: Faro / AEC Magazine

AEC Magazine reports that Faro launched Focus i.1 laser scanners and a mobile capture package aimed at faster colourised scanning and broader reality capture for AECO work.

The field implication is quicker production of point clouds and visual site records that can support verification, layout, progress and as-built review, provided registration and control remain trustworthy.

The product article does not establish field accuracy, capture speed across site conditions or a reduction in rework. Faster capture is not the same as accepted survey evidence.

Why it matters: Reality capture helps operations only when the crew can relate the record to the approved drawing, control points, date and specific decision it supports.

Practical AI use case or operational implication: Capture one workfront with the new workflow and compare registration, colour quality, control checks, repeat visits and reviewer corrections with the existing method.

Suggested executive takeaway: Require survey acceptance and revision/date metadata before a faster scanner output becomes a contractual, safety or payment record.

Actor: XGrids / AEC Magazine

AEC Magazine reports that XGrids unveiled the Lixel L3, a handheld device for true-colour point clouds, detailed meshes and photorealistic Gaussian splats.

The equipment is relevant to construction because a lighter capture workflow could make more frequent condition records practical in changing interiors, infrastructure corridors and handover inspections.

The article is a product announcement, not an independent accuracy or productivity trial. Colour and mesh detail do not establish survey control, completeness or fitness for a payment or safety decision.

Why it matters: Capture hardware becomes operational infrastructure when it reduces the cost of a reliable site record, not simply when its output looks more detailed.

Practical AI use case or operational implication: Compare one scan route with the current method, checking control, registration, missing areas, processing time and downstream usability in the model or issue system.

Suggested executive takeaway: Procure on verified accuracy, control workflow, export rights and maintenance burden; preserve the accepted record separately from the device output.

Actor: Matterport / AEC Magazine

AEC Magazine reports that Matterport Point Cloud Import enables firms to bring registered E57 files into Matterport's cloud platform.

The change can place survey-grade or specialist point-cloud data beside Matterport captures, giving project and operations teams a more continuous spatial view of existing conditions and installed assets.

The source describes an import capability, not a construction deployment or a guarantee that imported data will retain its control, coordinate system or metadata in every workflow.

Why it matters: Construction teams often have multiple capture instruments; a useful spatial platform must preserve the provenance and coordinate context of each source rather than flattening them into one visual.

Practical AI use case or operational implication: Import one controlled E57 into a project space and verify coordinates, registration, metadata, revision identity and export behavior against the survey record.

Suggested executive takeaway: Make coordinate-system, provenance and acceptance metadata mandatory before combining point clouds with project or asset decisions.

Actor: European Commission CORDIS / AutoConstruct-EU

The EU-funded AutoConstruct-EU project combines AI-driven mobile robots, multi-material 3D concrete printing and swarm coordination for on-site building and renovation.

The four-year project plans pilots covering a structural wall, a multi-storey interior partition and a renovation wall, with IoT-instrumented comparison against conventional methods and binders intended to replace more than half of clinker with supplementary cementitious materials.

CORDIS presents project objectives and targets, not completed field results. The equipment challenge is integrating navigation, printing, sensing, materials and fleet control on an unstructured site while preserving a qualified construction acceptance process.

Why it matters: Construction equipment is becoming a coordinated system of machines, materials and digital control rather than a single autonomous unit. That increases the importance of interfaces and recovery planning.

Practical AI use case or operational implication: For a robotics or 3D-printing pilot, map the material feed, machine state, site coordinate, human stop authority and acceptance check for each work package.

Suggested executive takeaway: Treat the project's 30% waste, 40% on-site CO2 and 25% time targets as hypotheses to validate, not as achieved performance.

Actor: University of Stuttgart / EIC Pathfinder

The University of Stuttgart said its SCALAR and COBRAS projects each received about €4 million through the EIC Pathfinder Challenge for autonomous robot collectives in dynamic construction environments.

SCALAR targets modular timber assembly with cranes and mobile robots; COBRAS targets robot swarms for reusable space-frame structures. Both projects use a co-design approach across architecture, engineering, robotics and computer science, with work scheduled to begin in November 2026.

These are funded research projects, not deployed jobsite systems. Their workforce significance is the multidisciplinary capability required to design, supervise, maintain and safely integrate construction robotics.

Why it matters: Construction automation creates a skills requirement around system integration and site judgment, not just a need for operators. Training and role design must precede autonomy.

Practical AI use case or operational implication: Map the roles a robot pilot would need—craft lead, controls engineer, safety lead, maintenance and data owner—and build a supervised training task around one assembly sequence.

Suggested executive takeaway: Treat research funding as a workforce signal; fund cross-disciplinary competence and site recovery skills before scaling autonomous equipment.

Actor: Bouygues Construction

Bouygues Construction opened Scale One, a 2,500-square-metre platform in Chilly-Mazarin designed to reproduce new-build and renovation conditions for full-scale testing before deployment on live projects.

The facility includes a FabLab, concrete 3D-printing area, robotics training space and test zones. Listed trials include AI-based waste management, Roby2 drilling robotics, lower-material construction, augmented wood and photovoltaic cladding.

The release is syndicated company material and does not provide independent performance results. The workforce lesson is practical: crews and managers can evaluate ergonomics, data and recovery before a technology meets a live project.

Why it matters: A physical testbed turns innovation from a demonstration into a training and acceptance process where site workers can surface failure modes before mobilization.

Practical AI use case or operational implication: Use a testbed-style gate for one new AI or robot workflow: train the crew, simulate conditions, record interventions and define acceptance evidence.

Suggested executive takeaway: Budget for hands-on training and full-scale validation; do not push an unproven system directly from demo to production site.

Actor: Dodge Construction Network, CMiC and Construction Dive

Construction Dive reports on a Dodge and CMiC survey of 235 U.S. general and trade contractors, with 85% expecting AI to reduce repetitive work and 75% expecting it to help teams learn from historical project data.

The report says 51% were actively evaluating AI changes and 40% had a dedicated AI budget. Large contractors were more likely than small and midsize firms to expect competitive advantage, while reliability and security remained major concerns.

The survey measures expectation and preparation, not productivity or safety. Its workforce relevance is the gap between interest and the training, governance and process ownership needed for responsible use.

Why it matters: AI adoption will depend on whether firms convert broad enthusiasm into role-specific competence and measurable changes in how estimators, project managers, superintendents and trades review work.

Practical AI use case or operational implication: Choose one repetitive role task, train users on source checking and escalation, and measure corrections, accepted decisions, cycle time and user confidence.

Suggested executive takeaway: Treat training, approved tools and help channels as adoption controls; do not turn survey optimism into staffing or authority changes without project evidence.

Actor: Green Badger

Green Badger announced Site Signal and Water Signal, tools that use project address and basic building inputs to prepare LEED v5 climate and carbon workbooks and calculate water-use baselines, savings and potential points.

The platform also enhanced Baili to identify products in uploaded construction submittals, match them to a verified green-products database and complete associated LEED documentation. Green Badger says the unified platform covers design through verification.

The release is vendor material and does not independently validate calculations, data freshness or certification acceptance. Public APIs and product matching still require a sustainability professional to review assumptions and evidence.

Why it matters: AI can reduce the administrative cost of sustainability work when it makes sources, methodology, assumptions and remaining actions visible instead of turning a calculation into an opaque score.

Practical AI use case or operational implication: Run Site Signal and Water Signal on one project, reconcile every input to official source data and compare the generated workbook with a sustainability professional's review.

Suggested executive takeaway: Use AI to prepare evidence, not certify it; preserve source dates, calculation assumptions, product records and qualified review in the project file.

Actor: IES / AEC Magazine

AEC Magazine reports the launch of IES Sync for Revit, intended to smooth the path between Revit building design and IES VE performance simulation.

The integration can help teams carry design information into energy and environmental analysis earlier, where envelope, systems and geometry decisions can be tested before they are expensive to change.

The source presents a connection workflow, not an AI-specific energy result or independent carbon reduction. Model mapping, assumptions, versioning and analyst review still determine whether a simulation is decision-grade.

Why it matters: Sustainability decisions become more useful when performance analysis stays connected to the evolving design rather than being recreated from a late export.

Practical AI use case or operational implication: Run one design option through the sync, compare geometry and system mappings, and record missing parameters, model changes and analyst corrections.

Suggested executive takeaway: Require a transparent model-to-simulation mapping and qualified review before energy or carbon outputs influence design approval or procurement.

Actor: Universidad Indoamérica / Frontiers in Built Environment

Researchers in Ecuador built a dataset of 330 images from 55 stabilized adobe blocks containing recycled PET, sugarcane bagasse fiber or Portland cement and trained a lightweight MobileNetV2 classifier.

The model achieved 80.95% held-out test accuracy on 42 images, with block-level cross-validation at 70.0%; the study used block-level splitting to avoid leakage and reports a 9.24 MB model suitable for offline commodity devices.

This is a laboratory baseline, not field certification or proof of production durability. The conservative cross-validation result and small physical sample make expansion, external validation and code acceptance necessary.

Why it matters: Sustainable materials need scalable quality evidence as well as lower-carbon inputs. An edge model could support inspection in resource-constrained production, but only after it survives real variation.

Practical AI use case or operational implication: Build a larger, block-separated field dataset across suppliers and conditions, then compare model findings with laboratory and qualified visual inspection.

Suggested executive takeaway: Treat the model as an inspection aid; require external validation, traceability and engineering acceptance before it influences material release.