Amogh N P
✦ In loving memory of Amogh N P — Architect · Designer · Visionary ✦
An industrial robotic arm assembling a complex curved computational timber lattice on a fabrication floor: the model's data driving the machine — file-to-factory made physical.
Unit VComputational Design Process

Fabrication & the AI Frontier

File-to-factory, mass-customisation — and AI as a co-pilot, not an architect.

≈ 45 min + studio task

The payoff of a computational model is that its data can drive a MACHINE directly. Learn file-to-factory (the model drives CNC, laser, robotic and additive fabrication), the process families, mass-customisation, and the BIM and digital-twin connection. Then AI/ML in computational design — its real uses (generative ML, surrogate models) AND its honest limits: training bias, no engineering judgement, 'AI image ≠ buildable design', authorship questions. The unifying ethic: the human stays the author of intent.

Learning objectives

By the end of this lesson, you will be able to — mapped to the course outcomes for Computational Design Process:

1
CO5 · Understand

Explain file-to-factory and the subtractive/additive/formative/robotic families.

2
CO5 · Understand

Explain mass-customisation and the BIM / digital-twin connection.

3
CO5 · Analyse

Describe AI/ML's real roles in computational design AND its honest limits.

4
CO6 · Evaluate

Keep the human as the author of intent — judge what computation and AI cannot.

The model drives the machine

File-to-factory & mass-customisation

The same logic that generates form generates the machine instructions; mass-customisation makes varied parts at standard cost; and BIM/digital-twin extend the data thread into operation.[1, 2]

File-to-factory model data toolpaths robotic arm CNC mill 3D printer The same logic that generates form generates the machine instructions — representation and production collapse into one.
DiagramFile-to-factory — the computational model's data drives the fabrication machine directly

The model drives the machine

FILE-TO-FACTORY (central to Kolarevic) means the geometry and fabrication data flow from the design model to the machine — CNC milling, laser/water-jet cutting, robotic arms, 3D printing — with minimal manual re-drawing. The same parametric logic that generates form also generates toolpaths, cut-lists and assembly data, collapsing the historic gap between representation (drawing) and production (building).[1]

Fabrication process families Subtractivemill / cut away Additivebuild up in layers Formativebend / shape Roboticgeneral-purpose Each has geometry rules (tool access, supportable overhangs) — designing FOR the process is itself computational.
DiagramThe fabrication process families — subtractive, additive, formative and robotic
A co-pilot, not the architect

The AI frontier — and its limits

AI accelerates generation and evaluation but lacks structural, code and constructability judgement — an AI image is not a buildable design. The human stays the author of intent.[4, 5]

AI — a co-pilot, not the architect AI accelerates ✓ • generative concept imagery• surrogate models (fast sims)• widens the option funnel but it lacks ✕ • structural / code / cost judgement• freedom from training bias• AI image ≠ buildable design the human stays the author of intent Computation, optimisation and AI are instruments of intent — never abdicate the design decision to a black box.
DiagramAI is a co-pilot that accelerates but lacks engineering judgement; the human stays the author of intent

Accelerating exploration

ML's legitimate roles: GENERATIVE ML (text/image-to-image and diffusion models for rapid concept imagery and options); ML-assisted optimisation / SURROGATE MODELS (a fast neural approximation of a slow simulation, so thousands of variants screen in seconds — genuinely useful in performance loops); classification (auto-tagging plans, flagging code issues); and generative layout for space planning. Used well, ML WIDENS the funnel of options and speeds the feedback loop.[4]

Static BIM vs digital twin

At a glance

AspectStatic BIMDigital twin
Data link to assetStatic BIM: noneDigital twin: live, sensor-fed
Time horizonBIM: design/constructionTwin: whole operational life
PurposeBIM: coordination, documentationTwin: monitoring, simulation, maintenance
UpdatesBIM: manual, by designersTwin: continuous, from real data
AI's roleCo-pilot: accelerates generate/evaluateNot: structural/code/constructability judgement
Vocabulary

Key terms

File-to-factory

Direct flow of model data to fabrication machines.

Additive manufacturing

Building form layer-by-layer (3D printing).

Subtractive manufacturing

Removing material from stock (milling, cutting).

Mass-customisation

Producing many varied parts at near-standard cost.

Digital twin

A live, data-linked virtual replica of a built asset (not just static BIM).

Surrogate model

A fast ML approximation standing in for a slow simulation.

Apply it

Studio task

Take one computational form and describe its file-to-factory path: which fabrication process you'd use, what geometry rule it imposes (tool access, supportable overhang), and how mass-customisation would label and assemble the unique parts. Then take an AI-generated design image and list three things that must be checked before it could be built — proving why 'AI image ≠ buildable design' and where you, the human, author the intent.

Check your understanding

Self-assessment

1. 'File-to-factory' means —

2. Why is 'AI image ≠ buildable design' the key caveat?

3. A digital twin differs from a static BIM model mainly because it —

In a nutshell

Recap

File-to-factory: the computational model drives CNC, laser, robotic and additive machines directly.
Processes are subtractive, additive, formative or robotic — designing for the process is itself computational.
Mass-customisation makes varied parts at standard cost — but labelling, tolerance and assembly become the new costs.
BIM carries data across the lifecycle; a digital twin adds a LIVE link to the built asset (static BIM is not a twin).
AI accelerates generation/evaluation but lacks engineering judgement — it is a co-pilot; the human authors intent.
The evidence

References & further reading

  1. [1]Branko Kolarevic (ed.), Architecture in the Digital Age: Design and Manufacturing (Spon Press, 2003) — file-to-factory.
  2. [2]Nick Dunn, Digital Fabrication in Architecture (Laurence King, 2012) — methods and implications.
  3. [3]Lisa Iwamoto, Digital Fabrications: Architectural and Material Techniques (Princeton Arch. Press, 2009).
  4. [4]Mario Carpo, The Second Digital Turn (MIT Press, 2017) — data/AI-era implications and authorship.
  5. [5]Current AI/ML tools and capabilities change rapidly — verify specifics against up-to-date sources; do not cite from memory.

Further reading

  • Branko Kolarevic — Architecture in the Digital Age.
  • Nick Dunn — Digital Fabrication in Architecture.
  • Mario Carpo — The Second Digital Turn.

Sources gathered and fact-checked June 2026. Published values vary by source, sample and method — treat as indicative and confirm against the cited standard before structural use.