Workflow / 01
Compare site options without losing the assumptions
A repeatable early-feasibility review that separates automated exploration from the decision to proceed.
AI / WORKFLOWS
AI architecture workflows
Practical guides for feasibility, layouts, visualization, BIM changes and survey boundaries, with deliverables and acceptance checks.
These are editorial pilot methods, not product integrations tested by GAD. Start with a familiar, non-confidential sample and measure corrections as well as generation time.
Workflow / 02
Turn a brief into a layout you can audit
Keep room requirements, adjacencies and human decisions visible through the move from written brief to editable plan.
Workflow / 03
Explore a visual direction without changing the design by accident
A controlled image-review method for material, atmosphere and client communication, with geometry checks at each acceptance point.
Workflow / 04
BIM-Edit: review AI changes across geometry, meaning and relationships
A June 2026 BIM benchmark offers a useful review lens: test the edited shape, object meaning and model relationships before accepting AI changes.
Workflow / 05
Gaussian splatting for adaptive reuse: define the capture acceptance test
A practical boundary between an immersive site record and measured project evidence, with capture planning, blind-spot checks and clear handover labels.
GAD / DESIGN STUDIES
Evaluate the output before adopting the workflow
GAD / DESIGN READING
FloorPlanCAD makes the line-level review matter
Use FloorPlanCAD research to plan a human-reviewed CAD symbol workflow, with instance checks, drawing boundaries and clear limits on quantity claims.
GAD / DESIGN READING
SceneScript turns a capture into a room hypothesis you can inspect
Review SceneScript as structured room reconstruction research, with explicit checks for wall relationships, openings, capture gaps and survey precision.
GAD / DESIGN READING
Energy surrogates need a shortlist check at the design target
Use an early-design energy surrogate to explore options, then verify shortlist ranking and target-adjacent errors in the underlying simulation.
GAD / DESIGN READING
SAM 2 facade masks need an explicit object boundary
Build a reviewable facade-image segmentation exercise with SAM 2, preserving original photos, mask corrections, exclusions and measurement limits.
GAD / DESIGN READING
AutoSpecs suggestions need a separate path to the specification
Review AutoSpecs submittal logs with clause-level evidence, separate AI suggestions, revision checks and a clear decision before items enter the project log.
GAD / DESIGN READING
Dynamo node recommendations still need a testable output
Use Dynamo machine-learning node recommendations with small facade-geometry tests, explicit list structure and repeatable checks before model changes.
GAD / DESIGN READING
FLOORA puts the floor plate into a checkable language
A generated residential floor plate becomes useful when a team can inspect its boundaries, understand its labels and decide what to change. FLOORA, an Autodesk Research preprint submitted on 28 September 2026, addresses that structured-output…
GAD / DESIGN READING
ARCH-B asks whether AI can recognise a building across drawings and photographs
An architectural assistant may describe a photograph fluently while linking it to the wrong plan. That error matters when a practice assembles precedent studies, checks image archives or prepares a client presentation. ARCH-B, submitted on 28…
GAD / DESIGN READING
IFCMemoryBench shows why project decisions need explicit sources
A BIM model can contain the geometry needed for a question while omitting the decision that gives that geometry meaning. A client may have changed a target in a meeting; a design team may have agreed which spaces to include in a comparison. An…
GAD / DESIGN READING
TestFit MCP turns a prompt into a constraint handoff
A natural-language site-planning request contains more decisions than it first appears to. “Fit the housing and keep enough parking” leaves the unit target, parking basis, access requirements and acceptable compromises unresolved. TestFit's MCP…
GAD / DESIGN READING
AI material suggestions need an assembly-level evidence trail
An AI-generated material shortlist is easiest to evaluate when the team can follow each suggestion back to a specific layer, requirement and source. A wall assembly offers a useful starting point: changing one component creates questions about…
GAD / DESIGN READING
BuildingNet: decide what a facade label means before counting it
A facade inventory sounds like a counting exercise until the team asks what counts as one window. Is a continuous glazed strip one component, several openings, or a collection of frames and panes? AI can accelerate classification only after that…
GAD / DESIGN READING
Scan2CAD: keep an aligned proxy separate from a verified asset
A clean library object can make an incomplete interior scan easier to read. It can also make uncertainty disappear too convincingly. Once a generic chair or cabinet sits neatly in the model, a later viewer may assume that its dimensions, identity…
GAD / DESIGN READING
SoundSpaces 2.0: turn an audio demonstration into a better acoustic brief
An architectural presentation can show a crowded reception space while leaving its sound entirely to the imagination. Adding an audio demonstration could make the discussion richer, provided the team knows what the demonstration represents. A…
GAD / DESIGN READING
Roof-GAN: use generated roof relationships to expose design decisions
A roof option can look coherent from above while leaving its most important design work unresolved at the intersections. Where does one plane meet another? Which edge becomes a valley? What must happen below the silhouette? These questions make…
GAD / DESIGN READING
Sat2Graph: check the connection before drawing the catchment
A single incorrect connection can change a site's apparent relationship to its neighbourhood. Two roads may cross in an aerial image while remaining separated vertically. A narrow break may conceal a real route. For early site analysis, these…
GAD / DESIGN READING
ShapeAssembly: ask which furniture changes a generated program can preserve
A furniture model becomes more valuable when a designer can change it without rebuilding its logic. Stretching a static object is easy; deciding which parts should move together is harder. An interior study often needs that second capability when…
GAD / DESIGN READING
Building-Diffusion: ask what the graph keeps fixed
A board covered with different building forms can suggest a broad design search. The more revealing question is which relationships were allowed to change. A change in visible form does not by itself explain which computational constraints were…
GAD / DESIGN READING
Point2Building: show where the context model completes the evidence
A complete neighbouring building makes a site model easier to use. It can also conceal how little of that neighbour was observed. An airborne scan may describe the roof more clearly than the facade, while trees and viewing angles leave other…
GAD / DESIGN READING
A damage classifier needs an explicit outside-the-test category
A classifier can only answer the question represented by its training and evaluation. That boundary becomes particularly important when the output concerns damaged buildings. A confident label may describe one of several studied damage patterns…
GAD / DESIGN READING
AI-filled weather gaps should remain visible in a design simulation
A simulation that runs without interruption is not necessarily using an uninterrupted record of measured weather. Missing observations may have been replaced before the file reached the design team. If those replacements are invisible, the…
GAD / DESIGN READING
Comfort AI should count the cost of asking
An occupant-centred building system can become surprisingly demanding of its occupants. Repeated questions about temperature may produce useful labels, but every request also interrupts someone. A model that learns efficiently should therefore be…
GAD / DESIGN READING
AI code checks need an unknown result, not a silent pass
A script can complete successfully while checking the wrong things. In a building model, an empty selection or missing parameter may produce a tidy report without answering the intended regulatory question. This makes execution success a…
GAD / DESIGN READING
OpenMask3D: reconcile the object list before planning reuse
An adaptive-reuse survey often begins with a practical question: which existing objects deserve closer inspection before the fit-out is removed? Searching a captured interior for chairs, cabinets or washbasins could help organise that work. The…
GAD / DESIGN READING
DiffusionLight: keep the lighting estimate attached to the montage
A proposed reception desk can look unconvincing in a photograph even when its perspective and dimensions are correct. Reflections point toward the wrong window; the shadow suggests a different light source. Lighting estimation offers a way to…
GAD / DESIGN READING
LayoutGPT: review the space a piece of furniture needs to work
A furnished room can satisfy a list of objects and still be awkward to use. A chair fits under a table but cannot be pulled back; a cupboard is inside the room boundary but its doors conflict with the bed. These problems suggest a useful review…
GAD / DESIGN READING
DINOv3: give visual precedent search an architectural question
An architect looking for a deep window reveal may receive a gallery of buildings with similar colours instead. The images can be visually persuasive while answering the wrong design question. Before introducing image-based AI into a precedent…
GAD / DESIGN READING
Infinigen Indoors: make synthetic survey data difficult on purpose
A room-recognition model may perform well on clear, spacious interiors and struggle when a chair hides a cabinet or repeated objects line a wall. Collecting more pictures of easy rooms does little to expose that weakness. Synthetic environments…
GAD / DESIGN READING
CityLearn: compare control strategies without hiding lost service
A building-control proposal may promise lower electricity costs while leaving an important question unanswered: what service changed to obtain the saving? A battery strategy, charging schedule or thermal-control policy needs to be assessed…
GAD / DESIGN READING
Text2CAD: review the design history of a generated component
A convincing model of a small architectural component can hide a difficult question: how would the designer change it? A shape that looks correct in one view may require rebuilding when its depth, opening or mounting relationship changes.…
GAD / DESIGN READING
TopoDiff: keep the load case attached to the structural image
An image of a branching structure can move quickly from a research figure into an architectural concept. The form may appear efficient because it resembles the flow of forces, but its meaning depends on the problem that produced it. TopoDiff…
GAD / DESIGN READING
TreeSatAI: keep a forest classification separate from a tree schedule
A green patch on an aerial image can represent a valuable landscape resource, yet it says little about which individual trees a building proposal would affect. AI species classification can help organize an early study of a large site. The…
GAD / DESIGN READING
NILMTK and the appliance evidence behind a retrofit brief
A household electricity bill can identify a large demand without explaining what an architectural retrofit should address. Appliance-level estimates may help a team decide which questions need better measurement before changing a kitchen, service…
GAD / DESIGN READING
MaterialGAN brings the physical sample into the rendering review
A material board often moves into an architectural rendering through a single photograph. That photograph combines surface colour, illumination and reflections in one image. MaterialGAN offers a research approach to separating aspects of surface…
GAD / DESIGN READING
FloodSformer makes the sequence of inundation a design question
A single flood-extent image can hide when water reaches an entrance, how long a service area remains affected and whether different parts of a site are exposed together. FloodSformer explores prediction of water-depth maps through time. For…
GAD / DESIGN READING
Docling and the hidden structure of an architectural schedule
An old door schedule can be readable on paper and still become unreliable when converted into searchable data. A number may survive extraction while losing the column heading or footnote that explains it. Docling is useful to examine at this…
GAD / DESIGN READING
Trajectron++ and the uncertainty at a public-space junction
At a museum forecourt, a person leaving the entrance may cross a cafe queue, turn towards a bus stop or pause to look for companions. A single arrow on a circulation plan compresses these different possibilities. Trajectron++ offers a research…
GAD / DESIGN READING
Learning-based nesting starts with the parts a workshop can cut
A tightly packed sheet layout looks efficient, but an architectural workshop needs every specified part to emerge with the right orientation, finish and cutting allowance. Learning-based nesting research becomes relevant when its geometric…
GAD / DESIGN READING
DeepSolar and the difference between an occupied roof and a solar opportunity
A campus energy plan can begin with a deceptively simple drawing: roofs with solar panels and roofs without them. DeepSolar makes that inventory question worth revisiting. Its architectural value lies in deciding where a closer investigation…
GAD / DESIGN READING
RampNet makes the missing curb ramp a survey question
An accessible entrance can meet a broken route at the edge of its site. The nearest bus stop may sit across two crossings, and one overlooked curb transition can determine whether the proposed arrival sequence works. RampNet offers a useful…
GAD / DESIGN READING
UV-Net and the search for a useful component family
A facade package may contain hundreds of brackets that look almost identical in a perspective view. Some share a machining sequence, some merely share a silhouette, and some must never be exchanged. Searching that collection is a plausible use…
GAD / DESIGN READING
Learning to Dispatch and the building sequence beyond the factory clock
A factory can finish a facade package quickly and still deliver the wrong pieces first. Corner units may be buried behind typical panels; completed cassettes may occupy the bay needed for the next operation. Learning-based scheduling is therefore…
GAD / DESIGN READING
Diffusion Policy and the architectural value of a small assembly trial
A construction detail is easier to understand when somebody must assemble it. The approach direction, gripping surface and final clearance become tangible rather than remaining implicit in a drawing. Robot learning research can sharpen that…
GAD / DESIGN READING
Building Data Genome 2 and the operating calendar behind an energy forecast
A building's operating calendar is part of its architecture in use. A library may extend evening access to one floor; a school hall may serve the neighborhood at weekends. Those changes alter the relationship between space, servicing and time. An…
GAD / REFERENCE LIBRARY
Research through architectural questions
116 source-linked records
Explore the collectionFOR ARCHITECTURE TECHNOLOGY BRANDS
Put a useful product in context.
Discuss a clearly labelled AI-column sponsorship or bilingual product feature. Editorial assessments remain independent.
Continue exploring architecture: Projects · Practices · GAD calculators & practical tools · Editorial standards
Questions to take into the work
How can an architecture studio pilot an AI workflow?
Choose one repeatable task with a reviewable deliverable. Preserve a baseline method, define acceptable errors and assign a person to inspect the result. Compare the output with the same brief and constraints, recording corrections and time spent. Expand only after a trial provides usable evidence.
What belongs in an AI design handoff?
Include the brief, input files, relevant constraints, tool and version information, output files and the human review decision. Mark unresolved geometry, semantic or coordination issues. Keep generated images separate from dimensional or model evidence so the next designer can see what has actually been checked.
Compare architectural AI tools · Architecture project studies · Tools
GAD Editorial · Answers reviewed 2026-10-18