Cabinet Shop Drawings With AI: Box Sizes, Reveals and Hardware Checks

Partly. AI can draft the schedules behind cabinet shop drawings (box sizes, door sizes, hardware lists), but don't trust it to hold a tolerance or to be the drawing you cut from. Before cutting, check box widths against the site measure, reveals against your gap standard and hardware clearances against the maker's current data, then build one unit first.

A shop drawing is the document the saw operator cuts from, so a figure that is 3 mm out costs a door or a drawer box. Vision models are weak exactly where drawings are demanding: OpenAI and Anthropic both say object counts are approximate, and both flag precise spatial localisation as a weak point. Treat AI as an assistant for schedules and arithmetic, and keep the drawing itself in your cabinet software or CAD.

Follow me on Instagram@sagnikteaches

The costliest errors come from hardware numbers that an AI recalls instead of reads. Blum's planning data for TANDEM runners with drawer sides up to 16 mm sets the internal drawer width at the internal cabinet width minus 42 mm, while TANDEM 19 mm, for sides 17 to 19 mm thick, says minus 49 mm. A model that remembers one figure and applies it to the other variant is 7 mm wrong, and the drawer binds.

Connect on LinkedInSagnik Bhattacharya

The worked run used throughout is one kitchen wall carrying five base units and five wall units, drawn at 3,400 mm and measured on site at 3,412 mm at worktop height. Its dimensions are invented for illustration.

Subscribe on YouTube@sagnikteaches

What AI drafts well for shop drawings, and what it cannot hold

The pattern is the same on every row. The AI produces tables and formulas, the truth sits in a measurement, a data sheet or your own shop standard, and a person reconciles the two. When nobody reconciles, the error reaches the saw. Dedicated AI drafting tools that promise finished shop drawings need the same routine, because the site measure, your gap standard and the hardware data are inputs only you hold.

If your cabinet software already produces the drawings, its hardware libraries are your first source of truth, and AI covers the paperwork around them. The tutorial on what AI adds to cabinet design software draws that boundary, and the joinery version of this question, can AI draw joinery shop drawings, covers the same ground for bespoke work.

From site measure to box widths: the dimension chain

A run is only as long as its narrowest point. Walls lean, so a reading taken at the floor says little about the width at the top of the wall units. Record the width between the two end walls at several heights, and tell the AI which heights each box occupies. In the worked run the readings are 3,418 mm at the floor, 3,412 mm at 900 mm (worktop height), 3,409 mm at 1,450 mm (the underside of the wall units) and 3,405 mm at 2,150 mm (their top). The site measure itself is a separate job, and the tutorial on a carpentry site measure checklist from notes and photos covers it.

The design drawing assumed 3,400 mm for both the base run and the wall run, with boxes of 600, 600, 800, 600 and 800 mm. The shop standard is a 10 mm scribe allowance at each end wall, so the run can be trimmed to the wall without touching a finished edge. That gives the prompt its rule: take the narrowest reading within the height the boxes occupy, subtract both allowances, and compare the result with the sum of the boxes.

Here is a wall run measured on site and the box schedule from the design drawing.
Readings are the width in mm between the two end walls at four heights: floor 3418; 900 mm high 3412; 1450 mm high 3409; 2150 mm high 3405.
Base boxes occupy 0 to 900 mm high: 600, 600, 800, 600, 800 (left to right).
Wall boxes occupy 1450 to 2150 mm high: 600, 600, 800, 600, 800.
Shop standard: 10 mm scribe allowance at each end wall.
For each run, use the narrowest reading within the height the boxes occupy, subtract both scribe allowances, and give me the available width, the sum of the boxes, and the spare or shortfall. If a run does not fit, suggest which single box to resize and by how much. Show every sum and say which reading each line used.

A first pass from a typical model reports that the base run is 8 mm short and the wall run is also 8 mm short, and suggests trimming the last box in each from 800 to 792 mm. Compare it with the right answer:

The base answer is right. The wall answer reused the worktop-height reading, so it understated the shortfall by 7 mm. A last wall box cut to 792 mm would have gone in 7 mm too wide at the top, and a fitter would have been planing a finished end panel on site.

The fix is a last box of 780 mm in both runs. The base run then totals 3,380 mm against 3,392 available, which leaves 12 mm spare. The wall run totals 3,380 against 3,385, which leaves 5 mm. Five millimetres is thin, and whether it is acceptable is the maker's call, not the AI's. If you want 10 mm spare on the wall run too, take the other 800 mm wall box down to 795 mm, or plan a filler.

Put the rule into the Project instructions as a sentence: "For any run, use the narrowest reading within the height the boxes occupy, and say which reading each line used." The second half is what lets you spot a wrong reading in ten seconds.

Reveals: a gap budget for doors and drawer fronts

A reveal here means the visible gap or margin left around a door or drawer front. Treat it as a budget that you spend across each run and each stack, with the same rule every time. The standard in the worked shop is 1.5 mm from a door's edge to the edge of its unit, so neighbouring doors sit 3 mm apart across a unit joint. It is 3 mm between the two doors of a pair, 3 mm between stacked drawer fronts and 1.5 mm at the top and bottom of a stack. Yours will differ with hinge type and finish. What matters is that the standard is written down in the Project.

Door widths follow from it. A single door on a 600 mm unit is 600 − 1.5 − 1.5 = 597 mm. A pair on an 800 mm unit is (800 − 1.5 − 1.5 − 3) ÷ 2 = 397 mm each, and on the resized 780 mm unit each door of the pair is 387 mm.

The first-pass column comes from a model that read "3 mm reveal" as a figure per edge. Each unit's doors come out 3 mm narrower in total, and neighbouring doors sit 6 mm apart across each joint, double the standard. The arithmetic within each unit still closes, so adding the widths up will not expose it. What exposes it is a second table listing every gap along the run, with its position and size. A joint gap of 6 mm against a standard of 3 mm is obvious the moment it is written down.

The same budget runs vertically. A three-drawer pack with a front zone of 720 mm has 1.5 mm margins at the top and bottom and two 3 mm gaps between fronts, which leaves 711 mm for the fronts themselves: 141, 285 and 285 mm. The AI's first answer was 140, 290 and 290, which add up to exactly 720 and leave no gaps at all, so the fronts would touch. A one-line check catches it: fronts plus gaps must equal the zone height.

Hardware clearances come from the data sheet, never from memory

Blum's planning data for the TANDEM runner gives each variant its own width rule. SKW is the internal drawer width and LW the internal cabinet width. For TANDEM with drawer sides of 11 to 16 mm, SKW = LW − 42 mm. For TANDEM 19 mm, which takes drawer sides 17 to 19 mm thick, SKW = LW − 49 mm. Both rules carry +0.0 / −1.5 mm beside them, and the drawer length is SKL = NL − 10 mm, where NL is the nominal length of the runner. These figures come from Blum's 2024/2025 catalogue and technical manual, so check the current planning data for your runner's part number before relying on any of them.

Take a 600 mm base unit with 18 mm carcass sides, so LW is 564 mm:

The third row is the typical AI error. The model remembered 42 from the thinner-side variant and applied it to a box with 19 mm sides. The box comes out 7 mm too wide, at 560 mm instead of 553 mm, and each side keeps 2 mm of running space instead of 5.5 mm. Nothing on the drawing looks wrong, because 522 mm is a perfectly plausible internal width. The drawer rubs, and nobody knows why until the first box is built. For a 500 mm runner the length rule gives a drawer 490 mm long.

Make the AI read instead of remember:

Use only the attached data sheets (blum-tandem-16.pdf and blum-tandem-19.pdf). For each drawer box type in the schedule give: the formula used, the page it came from, the internal cabinet width, the internal drawer width, the outside width for my drawer side thickness (state the thickness you used) and the gap each side.
If a figure is not in the sheets, write NOT IN SHEET instead of using what you remember. Quote each sheet's edition or date.

Save the planning pages for the runners you use as a short PDF. A few pages sit well inside Claude's 100-page limit for visual analysis of a PDF, and ChatGPT reads both the text and the drawings in an attached PDF too. Hinges follow the same logic. The hinge maker's data gives the cup position, the drilling distance from the door edge and the smallest gap each overlay type allows. The common 35 mm cup hole is standard, but depth, edge distance and gap limits differ from hinge to hinge. Appliance cut-outs come from the installation sheet for the exact model, because two models of the same nominal width can state different openings.

Door swing is a clearance too. At a corner, a door that opens against a neighbouring handle or appliance needs space for both the door and the handle projection. Ask the AI to list every door that opens against another door, handle or appliance, then test each on a mock-up. It can flag candidates, but it can only judge the clash if you give it the handle projection figure.

Box sizes in a spreadsheet: formulas from AI, tested on three units

Models are better at writing a formula than at doing a long sum reliably, and a spreadsheet shows every step. Ask the AI for a sheet in which the inputs sit in labelled cells and everything else is a formula, then audit it cell by cell. A drawer-box sheet needs only a handful of cells:

B2   Cabinet width W (mm)                   600
B3   Carcass side thickness T (mm)           18
B4   Runner subtraction S (from the sheet)   42
B5   Drawer side thickness DS (mm)           16
B7   Internal cabinet width LW               =B2-2*B3
B8   Internal drawer width SKW               =B7-B4
B9   Outside drawer width                    =B8+2*B5
B10  Gap each side                           =(B7-B9)/2

Test it on three units whose answers you already know from the table above. A 600 mm base with 16 mm sides should give 522, 554 and 5.0. An 800 mm base should give 722, 754 and 5.0. A 600 mm base with 19 mm sides and S set to 49 should give 515, 553 and 5.5. If any result differs, the formula is wrong, not the unit.

Type S and DS from the data sheet and write the sheet's edition or date in the cell beside them. When the maker issues a new edition you change one cell and every box follows. The ChatGPT Excel guide explains how to have AI write, explain and debug formulas like these, and the same method applies to a shop's box sheet.

Five ways the draft goes wrong, and how each one showed up

1. The wrong reading for the height. The wall run reused the worktop-height measurement and understated the shortfall by 7 mm. It showed up when the output was made to name the reading each line used.
2. A gap counted per edge instead of per joint. Each unit's doors came out 3 mm narrower in total, with 6 mm between neighbours. It showed up in a table listing every gap along the run.
3. A stack with no gaps in it. Drawer fronts of 140, 290 and 290 mm filled 720 mm exactly. It showed up when fronts plus gaps were added back to the zone height.
4. A hardware figure recalled, not read. The 42 mm rule was applied to 19 mm drawer sides. It showed up on the first test box, which rubbed.
5. A unit slip in the input. One reading was typed as 341.2 from a tape marked in centimetres, and the AI took it as millimetres. The available width came out at 321 mm against boxes totalling 3,400 mm, an absurd shortfall that gave the game away. Make the AI state the unit for every number it uses, and flag any reading outside a plausible range for the room.

Four of those five fall to a check that takes a minute. The hardware error is the one that can slip through on paper, which is why the routine below ends with a built box.

The pre-cut routine, from site measure to first-off box

1. Measure and record. Take readings at set heights on a printed sheet, photograph each wall with the tape in view, and write the unit beside every figure.
2. Build the schedule. The AI drafts the box table. You re-add it against the readings, using the narrowest reading for each height.
3. Apply the gap standard. Door and front sizes come from the AI, along with a table of every gap and the check that fronts plus gaps equal the zone.
4. Read the hardware. Formulas come from the current data sheets, each with its edition and page, and NOT IN SHEET stands in for anything missing.
5. Build first-off. Make one box of each type: a door unit with its hinges hung, and a drawer unit with its runner fitted and a drawer slid in and out. Try each handle against its neighbour.
6. Sign it off. The person who will fit the kitchen initials the drawing, and any revision gets a new number so the saw never cuts from an old one.
7. Cut the batch. Only now.

If any check fails, correct the source rather than the number: the gap standard, the formula or the reading. A patched number comes back on the next job, while a corrected source does not. A first-off box costs about an hour of shop time, while a batch of wrong drawer boxes can cost a day and a stack of board. The same schedule also feeds the price, which is why the cabinet maker quote template uses the unit table you build here.

What the client sees, and why an AI picture is not a drawing

Clients approve what they can read, so give them elevations with the key dimensions, the door style and finish, the handle position and the worktop edge. AI image tools can produce attractive pictures of a kitchen, but those pictures invent details such as hinge sides, handle heights and panel joints, and nothing in them is dimensioned. Label any such image "concept only" and keep it away from the approval drawing, otherwise the client approves a picture the shop cannot build.

Keep a revision log as well. Every change after approval, whether a unit moved, a door handing swapped or an appliance model changed, gets a revision letter, a one-line description and a date, and the AI can draft that log from your notes. When the drawing changes, the schedules, door sizes and hardware counts all need re-running, so ask the AI to list what each revision touches and re-run those steps instead of patching one number. Once the drawing is signed off, the tutorial on turning a signed-off drawing into a cutting list and board order picks up from there.

Further reads

• Cabinet Shop Software: Quotes, Cut Lists and Orders, and What AI Adds — Compare shop software that links drawings to quotes, cut lists and orders.
• Cutting List Software vs AI: Which Gives a Carpenter a Reliable Cut List? — Decide whether cut-list software or AI should produce the cutting list.
• AI Software for Construction Drawings: Catch Changes Between Revisions — Catch changes between drawing revisions before they reach the saw.
• Joinery AI App Checklist: What to Check Before You Pay for One — Run a checklist before paying for any joinery AI app.
• How Does AI Help in Cabinet Making? Six Workshop Jobs Explained — See the six cabinet-making jobs where AI helps most.
• AI Tools for Shop Drawings: Checking Fabrication Drawings Against the Spec — Check fabrication drawings against a written specification.
• Cabinet Maker AI: Answer Showroom Enquiries and Book Site Measures — Triage lanes, an enquiry form, a rules sheet, a booking page and reminder texts for a cabinet maker, with a worked month of 56 enquiries and what it costs.
• How to Use AI in Carpentry: Cut Lists, Takeoffs and Client Quotes — A fitted wardrobe worked end to end: AI cut list with formulas, seven hand checks, wastage decision, order list and quote wording a client can hold you to.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Blum's catalogue and technical manual 2024/2025 (TANDEM planning pages for 11-16 mm and 19 mm drawer sides) and Blum's TANDEM product pages; OpenAI's image input guide and file-upload help page; Anthropic's vision documentation and its help article on uploading files. Dimensions in the worked run are illustrative.