Agents for Schedules & Quantities

Extracting schedules and quantities is close to the ideal agentic task: it is exhaustive, mechanical and rule-based, and it operates on data the model already holds. A BIM model, in principle, knows every door, its type and its dimensions; every room, its name, number and area; every wall, floor and finish and its extent. An agent that can read the model and act on it can turn that latent data into a door schedule, a room-and-finish schedule, or a quantity take-off far faster and more consistently than a person clicking through elements - and it can regenerate the whole thing in seconds when the design changes, keeping the schedules live rather than stale.

The same holds, more roughly, for drawings without a rich model. Agents that can read drawing sets can count and measure from them - tallying symbols, reading dimensions, estimating areas - which matters enormously in the many Indian projects and small practices that work in 2D CAD or even on drawings rather than a full BIM model. The counting is less certain from drawings than from a clean model, but the labour it saves is even greater, because manual take-off from drawings is one of the most tedious jobs in the office.

Here is the pivot the whole lesson turns on: the agent does not know the building; it knows the source you point it at. Its count of doors is a count of the doors in the model or on the drawings - not a count of the doors the building needs. If the model is incomplete, the count is incomplete; if an element is mismodelled, it is miscounted; if the drawings are ambiguous, the reading is a guess. This is the oldest rule in computing, garbage in, garbage out, and it bites hardest exactly where the output is a confident total that hides its own gaps. So the first discipline of agentic take-off is not about the agent at all - it is about the source: a quantity is only as trustworthy as the model or drawing it was extracted from, and reconciling the count begins with knowing the state of what was counted.

To reconcile a count you have to know how it fails, and agentic take-off fails in a handful of characteristic ways - each of which reads as a normal, confident number. Knowing the taxonomy turns verification from a vague unease into a checklist.

Source errors come from the model or drawings, not the agent: an incomplete model, mismodelled elements, ambiguous drawings, objects on the wrong layer or of the wrong type. The agent faithfully counts what is there, so its count inherits every flaw in the source. Rule errors come from the measurement method: quantity take-off is governed by conventions - what to include, how to measure an area, whether to net out openings, how to handle waste - and an agent left to choose a rule will pick a plausible one that may not be yours or your standard method of measurement. A count done to the wrong rule is precise and wrong. Counting errors are the classic ones: double-counting elements that appear on multiple views, missing items that fall outside the query, miscategorising an element so it lands in the wrong schedule line. Arithmetic and unit errors - a factor of a thousand, a square-metre/square-foot slip, a summation mistake - are less common in a tool-using agent that computes rather than guesses, but they happen, especially when the agent reasons in prose rather than calling a reliable calculation.

The most dangerous category, though, is the silent gap: the thing the agent was never asked to count and so did not. An agent told to schedule doors will schedule doors; if there is a category of opening it did not recognise as a door, that category simply does not appear, and the schedule looks complete. A missing line is far harder to catch than a wrong one, because nothing on the page draws your eye to it. This is why verifying a schedule is not just checking the numbers that are there, but asking what should be there and is not - a question only someone who understands the building can answer, which is precisely why the count cannot be trusted to the agent alone.

Reconciliation is the craft of turning a fast agentic count into a number you can stand behind, and it is a specific, learnable set of moves rather than a re-count by hand (which would throw away the whole benefit). The goal is high confidence at low cost, using the fact that errors leave traces.

Start before the count, with a model or source audit: is the model complete for what you are about to count, are elements correctly typed and placed, are there known gaps? A count from an audited source is worth ten from an unknown one. Give the agent the measurement rule explicitly - the method of measurement, what to include and exclude, how to treat openings and waste - so a rule error cannot creep in through the agent's default. Then, on the output, apply cheap high-leverage checks. Sample counting: pick a room or a floor, count one element type by hand, and compare to the agent's figure for that subset; a match builds confidence in the whole, a mismatch localises the problem. Order-of-magnitude sanity: does the total make sense for a building this size - is the tile area roughly the floor area, is the door count plausible for the room count? Gross errors announce themselves to a designer's intuition. A second method: extract the same quantity a different way - a different query, a different tool, or a rough manual estimate - and reconcile the two totals; independent counts that agree are strong evidence, and a divergence is a flag worth chasing.

Crucially, weight reconciliation by consequence. A schedule used to check the design is low-stakes; a take-off feeding a tender, a purchase order or a client budget is where an error becomes money, and deserves real reconciliation. And build in the human question the agent cannot ask itself - what is missing? Walk the building in your mind and ask whether every category that should be scheduled is present. Reconciliation is not distrust of the agent; it is the professional step that converts a quick, confident, possibly-flawed total into a defensible basis for a decision - and it is far cheaper than the re-count it replaces, because you are checking a number the agent produced rather than producing it yourself.

Schedules and quantities feel like neutral, factual output - just counts - which makes it easy to forget that they are among the most consequential numbers a design office produces. A quantity take-off becomes a cost estimate becomes a tender becomes a purchase order becomes what actually gets built and paid for. An error in the count is not a documentation blemish; it is a budget blown, a tender mispriced, a material short-ordered on site, a client's trust lost - and, like everything else in this module, it lands under your name, not the agent's.

That is why the line here is drawn at consequence. Use agents freely to generate schedules and take-offs, to regenerate them as the design changes, to do the exhausting counting that used to eat days - the productivity is real and the tedium relieved is real. But the moment a number is going to be relied upon for a commitment - a cost, an order, a contractual quantity - it must pass through reconciliation and become yours. An unreconciled agentic take-off handed to a client as a budget, or to a contractor as a tender basis, is a liability dressed as a spreadsheet: it will look authoritative, it may be materially wrong, and the fact that an agent produced it is no defence when the money runs out.

There is also a quieter professional point. Quantities connect to cost, and cost connects to the client's interest and often to statutory and contractual obligations; a designer who lets an agent's count flow unreconciled into a budget has not saved time so much as transferred risk onto the project and their own liability. The mature use is the opposite: let the agent carry the counting, spend the time you saved on reconciling the numbers that matter, and present figures you have actually stood behind. That is the whole shape of agentic documentation in one task - the agent does the tireless, exhaustive work; you supply the judgement about what is trustworthy enough to build and buy from; and the number that leaves your office as a commitment is one you have made your own.