On a large construction project, the schedule is a political document. It says who has to deliver what and by when, and it becomes, often against the intention of those who wrote it, the ground on which claims are built.
Running that schedule is the role France calls OPC (scheduling, control and coordination), and it has been served by the same software since the mid-2000s. MS Project, Primavera P6 and their cousins (Tilos, Asta Powerproject, Plexos) are still here, reliable, and they are the ones you find on large operations. And that is fine.
The problem is not the tool. The problem is what the tool cannot do, and what scheduling teams have to make up for by hand, every week.
Four recurring blind spots
1. Prediction has stayed manual
All these tools can compute a critical path, identify float and simulate an alternative scenario. None of them can, out of the box, anticipate that a subcontractor is about to slip before it slips. Prediction is left to the scheduler, who has to build it in their head from the history of delivery notes, correspondence and site reports.
The consequence: drift is documented after it has happened, not before. The up-to-date schedule is in reality a record of where things stood a fortnight earlier.
2. The schedule does not talk to anything else
The site manager's email announcing a site instruction received on Friday evening. The architect's note on a technical detail that has changed. The foreman's photograph of a defect in the foundations. All the information that ought to feed the schedule lives somewhere else: in Outlook, in the document management system, in WhatsApp, in someone's head.
The modern scheduler spends a good part of the week rebuilding scattered information in order to project it into the schedule. That is translation, not analysis.
3. The as-planned against as-built picture is invisible
A baseline schedule is signed at the start of the project. Successive amendments change it, on paper, sometimes in the tool. But the schedule as it should have been kept and the schedule as it is actually kept are almost never set side by side.
The result: when the time comes to defend a contractual position, only part of the facts is documented, and the rest is reconstructed under pressure in the weeks before the claim. It can be done, it is done, but it is expensive and it is fragile.
4. Reporting is badly calibrated
A Gantt chart twelve levels of nested tasks deep, shown to an executive board that has fifteen minutes, says nothing. A client dashboard reduced to three indicators does not say much more. Between the two lies a job of prioritising that the tool does not do on its own.
A schedule is not caught up at the end of a meeting. It is either kept, or reconstructed, and reconstruction is expensive.
What AI can concretely do
To be clear from the outset: there is no question of replacing MS Project or P6. Those tools remain the base: a reference format, an exchange standard, a calculation framework. What can be added is an intermediate layer that addresses the four blind spots above.
Continuous reading of the sources outside the schedule
A local agent that reads the coordination emails, the minutes and the field notes, and automatically identifies everything that ought to touch the schedule. Not in order to change it, in order to tell the scheduler that a decision is waiting.
For example: a site instruction received on Thursday changes the sequence between the structural works and the MEP packages. The agent reads it, alerts the scheduler and proposes two ways of taking it into the schedule. The scheduler decides, and the schedule is updated on Friday morning, not three weeks later.
Predictive simulation on weak signals
Cross the schedule, the delivery notes, the hours declared and the correspondence to spot the weak signals. A subcontractor who asks three times in one week about a technical detail is rarely signalling that all is well.
The aim is not an exact prediction. It is an early warning, which turns an undetected drift into a team discussion two weeks before it becomes a claim.
The as-planned against as-built kept automatically
A continuous comparison between the baseline schedule (signed), the current schedule (amendments included) and the observed schedule (what actually happened on site). The gap is computed all the time, not rebuilt the night before an expert meeting.
The benefit is not a contractual innovation: it is simply that the defence or claim file is ready at any moment, because it writes itself continuously.
Reporting fitted to the audience
The same schedule, read differently depending on who receives it. For the contract manager, the sequence of site instructions and their contractual impact. For the client, the milestones committed and the float remaining. For the executive board, three indicators: on track, tight, drifting. Not three manual exports: one source, several generated views.
Why local, and not another SaaS
The temptation, in 2026, is to offer a US SaaS that would do all of this, billed per seat per month. It already exists, and it is moving fast. But when the subject is the schedule of a sensitive operation, a live claim, a contractual position, handing that data to a third-party supplier is a heavy choice, and it deserves to be weighed.
The alternative Cabestian offers with its engine is deliberately different: an agent installed on the machine at the site, which reads your sources on your premises and works with your scheduling team without sending anything outside. You keep ownership, readability, and the way out.
A closing word
MS Project, P6 and Tilos are finished tools within their scope: they do what they do, and they do it well. We are not going to ask more of them. But the gap between what they allow and what a modern scheduling team should be able to do has grown so wide that it is worth instrumenting.
That is what we install: project by project, without a revolution, without imposing a new tool on the teams. The intermediate layer lives beside the standard schedule, completes it, and disappears if it does not hold.
If this rings true, if you recognise the four blind spots in your own scheduling today, there is probably something to discuss. Booking a call is there for that: a few questions, and we set up a conversation.