Construction scheduling with Claude: check dependencies before the Gantt

Prepare a construction schedule with Claude →

On a renovation site, the schedule shows a two-day overlap between the electrician and the plasterer, in the same room, the same week. Nobody noticed before the site meeting, because the Gantt chart had been built trade by trade, without re-checking dependencies between crafts. The resulting two-day delay then adds to every following task, until it represents nearly two weeks by the end of the site. A readable Gantt chart had nonetheless been produced; that was not what was missing.

A construction schedule is not worth its clean bars, but the soundness of the constraints that explain its dates: dependencies between trades, actual availability, known closures and supply uncertainties. A sequence drawn without this data remains a drawing, not a management tool.

This article details what an automatically generated Gantt chart can and cannot establish, the types of dependency between tasks and their respective risks, the method for building a revisable schedule without losing the assumptions set, and the situations where the schedule alone is not enough.

Summary

1. A bar on a Gantt chart is not a guarantee of a deadline
2. Dependencies between tasks: four types, four different risks
3. Building the schedule step by step
4. Revising without losing the assumptions set
5. When a construction schedule is not enough
6. What we observe among architects and construction firms
7. Training to make scheduling more reliable
8. Frequently asked questions

A bar on a Gantt chart is not a guarantee of a deadline

An automatically generated schedule shows dates consistent with the rules it was given; it does not guarantee those rules match the reality of the site. The document's reliability depends entirely on what was checked before it was generated, never on the visual quality of the result.

Distinguishing an estimated duration from a confirmed lead time

A default task duration, or one carried over from a previous site, remains an estimate until the relevant contractor confirms it. A supply lead time unknown at the time of planning must appear as such on the document, never as a firm date chosen simply to fill an empty cell.

The generated workbook stays static

A Gantt chart produced in Excel by a script does not automatically recalculate following tasks when a date is manually changed in the file: every constraint change must be carried into the source configuration, then the document regenerated from it. Changing a single cell without going through that step leaves the rest of the schedule silently out of date.

Dependencies between tasks: four types, four different risks

A dependency between two tasks is not simply "one follows the other": its exact nature changes what must be checked before the schedule can be considered reliable.

An unchecked overlap costs more than an isolated delay

The first day of the site determines the rest of the calculation

The schedule's first week generally corresponds to the calendar week, Monday to Sunday, that contains the chosen start date — a detail that must be checked in the Gantt headers before reading any later date, since a one-day error at this point propagates through every task that depends on it.

Building the schedule step by step

Step 1: Gather tasks, trades, known durations and imposed milestones. This initial list must explicitly separate what a contractor has confirmed from what remains a working assumption.

Step 2: Identify each dependency and its exact type. A misclassified dependency — an overlap treated as a simple succession, for example — then distorts every date calculation that relies on it.

Step 3: Flag closures and uncertain supplies before generating the file. An unknown supplier lead time must appear as an explicit unknown on the schedule, never as a date arbitrarily chosen to fill the calendar.

Step 4: Generate the file from the configuration, never by editing cells by hand. This step ensures that following tasks actually incorporate every constraint set upstream.

Step 5: Compare milestones with bars before circulation, then validate availability with participants. A schedule not confirmed with the contractors involved remains a draft, not a shared commitment.

Revising without losing the assumptions set

Every schedule revision should produce an explicit list of affected tasks, dates before and after the change, and assumptions still awaiting confirmation — never a new file silent about what changed and why.

A moved task changes more than its own duration

Shifting a task by a few days can change a shared access point, a scheduled delivery or equipment availability, even when its own duration stays unchanged. These indirect effects are exactly what a systematic review of the schedule, after every revision, catches before they turn into a new delay on site.

Keep every dated version with its configuration

The configuration file that produced a given version of the Gantt chart must be kept alongside it, or the ability to reconstruct, weeks later, the assumptions that justified a now-overdue date is lost. Decisions made at a site meeting are best documented separately in minutes, which the schedule then translates into timing consequences.

When a construction schedule is not enough

A properly built schedule does not resolve every situation that can arise during a site.

A contract providing for contractual delay penalties, whose principle falls under the clause pénale defined by Article 1231-5 of the French Civil Code, requires an analysis specific to the contract concerned before drawing any consequence from a shifted deadline: a revised schedule does not on its own determine whether a delay is attributable to a contractor, nor whether such a clause actually applies. A persistent conflict between two contractors over access to the same zone is likewise not resolved by a new Gantt chart, but by arbitration organized at a site meeting. Finally, a major unforeseen event — prolonged bad weather, an unexpected discovery during works — requires rebuilding the schedule's assumptions before any new generation, rather than mechanically shifting every task by the same number of days.

What we observe among architects and construction firms

In our exchanges about site scheduling, the difficulty is almost never the logic of dependencies itself: a site manager knows that paint waits for a dry coating. What costs time is reformatting the Gantt chart after every meeting — retrieving the previous configuration, checking that a one-off change has no cascading effect on distant tasks, or simply reconstructing assumptions set weeks earlier. That is repetitive update work, not a lack of skill — and it is exactly the kind of task a well-structured script removes, without ever deciding on the site manager's behalf whether a given constraint is actually being met on the ground.

Training to make scheduling more reliable

Setting up a schedule template once and for all is not enough: it also needs to be kept alive from one meeting to the next without losing the assumptions set upstream. For an architect practising as a self-employed professional (NAF code 7111Z in France), the FIFPL fund covers part of the cost of Qualiopi-certified training under 2026 criteria set at €300 per day and €900 per year, with e-learning capped at 50% of the daily rate. This funding must be arranged before training starts, not after.

Frequently asked questions

Does an automatically generated schedule guarantee deadlines will be met?

No: it displays dates consistent with the constraints it was given, but guarantees neither their accuracy nor that they will be met on site. Its reliability depends entirely on the quality of the durations, dependencies and availability checked before it was generated. A schedule not confirmed with the contractors involved therefore remains a draft calendar, not a contractual commitment to the dates shown.

Why is editing a date directly in Excel a problem?

Because the generated workbook stays static: a manually edited date does not automatically recalculate the tasks that depend on it, leaving the rest of the schedule silently inconsistent with the change made. Good practice is to carry the change into the configuration file, then regenerate the document. Doing so also keeps an explicit trace of what changed and why, rather than a silent edit that would be impossible to reconstruct weeks later.

How should an overlap between two trades in the same zone be handled?

An overlap must be explicitly authorized by a checkable condition — an organized shared access, for example — before being entered as such on the schedule. Without that condition, the overlap should be treated as a conflict to resolve, not as a simply acceptable simultaneity. That conflict should be flagged explicitly before the document is generated, never left for the contractors to sort out on site the day both crews arrive in the same spot.

What happens if a delay triggers a contractual penalty?

Applying a delay penalty, whose principle falls under the clause pénale set out by Article 1231-5 of the Civil Code, depends on the terms specific to the contract concerned and on whether the delay is actually attributable: a revised schedule documents the shift in dates, but does not settle this question on its own, which calls for a contractual and often legal analysis. The schedule nonetheless provides a useful piece of evidence for that analysis, by precisely dating when the delay appeared and which tasks it affected. Determining whether that delay is attributable to a contractor rather than an external event remains a separate question, to be handled apart from the calendar itself.

Does the entire schedule need regenerating after every small change?

Yes, as soon as a date constraint changes: regenerating the document from the updated configuration ensures every dependent task actually reflects the change, whereas an isolated manual edit leaves the rest of the file out of date without this being immediately visible. This systematic regeneration stops a distant task, indirectly dependent on the changed constraint, from keeping a now-incorrect date unnoticed until the next meeting. It also ensures the version circulated to contractors is always the one produced from the latest validated configuration.

A reliable construction schedule rests on three elements that do not change from one site to the next: durations and dependencies checked before generation rather than assumed, systematic regeneration from the configuration after every change, and traceability of assumptions kept with every dated version. Once that foundation is in place, it is exactly what the construction-schedule generator applies automatically, while leaving the site manager the decision on whether constraints are actually being met.

Decisions that change the schedule are usefully documented in site meeting minutes, and a defect discovered during the site is best tracked in our defect-tracking register guide through to its resolution.

Training 100% fundable via OPCO/FIFPL. Qualiopi-certified programme. To structure your schedules and site documents with AI as part of our AI for architectural management training, contact Educasium and specify your status (employee, self-employed, business owner) and your goal.