The technical director of Berlin’s new airport told a parliamentary inquiry about one kind of visit. Some of the airport’s customers, he said, made up their minds too late. They would walk round the building site, stop at a column and say they didn’t want it there.
That column has been in the plans for three years, he would reply.
Then you’ll have to take it out.
That can’t be done.
By his account, the freeze on design changes was lifted once again: now we have to change it after all, because the customer wants it.
The terminal had been changing for years. A walk-boarding pier was added in 2006, partly because low-cost airlines preferred walking to the plane over riding a bus. A second pier followed in 2007, along with a bigger shopping area to help pay for the airport. In 2008 a full level was fitted out for non-Schengen passengers, and double-deck jet bridges were added. In 2010, new EU security rules led to two extra pavilions. Several came with their own price tag. In July 2008 the supervisory board approved €18.6 million for the non-Schengen work and €24.6 million for the jet bridges. Next to a construction budget of about €2.4 billion, each looked small.
In January 2010, right after approving a batch of change requests, a management meeting minuted: “Management reaffirms the freeze on planning changes. It is to be strictly observed.”
It wasn’t. In June 2010 the board moved the opening by seven months, to June 3, 2012, on the condition that no more significant changes were made. In January 2012 the site supervisors reported about 130 new changes from 2011 alone that still had to be fitted into the plans. In May 2012, 26 days before the first flights, the opening was called off. The airport opened in October 2020. The case has the rest.
Every change had a reason. Taken together, the inquiry found, design changes were usually at least one of the main causes of the delays.
What creeps
A project’s scope is what it has agreed to deliver. Scope creep is that list growing while the time, money and people stay the same. A 2009 paper for the Project Management Institute’s global congress quotes the PMBOK Guide, the institute’s handbook: “adding features and functionality (project scope) without addressing the effects on time, costs, and resources, or without customer approval.”
The paper is careful about what doesn’t count. Change on a project is normal, and “if an expansion of scope is approved, then it is not scope creep.” The trouble is the addition whose real cost nobody weighs.
This page uses the handbook’s sense: did anyone deal with an addition’s effects on time, cost and resources? The pavilions were approved. But by the testimony of the board’s chairman, Klaus Wowereit, the board was never told their full knock-on effects. He told the inquiry that when management proposed an extension, the board assumed management had already checked what it would disrupt. Had those effects been laid out, he said, the board could have asked whether it was worth the risk. “But it wasn’t presented.” Had he known it would hold up everything, he said, he would rather not have built it.
Additions are hard to weigh because their cost lands somewhere else. At Berlin, the technical director told the inquiry that a single change request touched every trade on the building. The general planners’ representative, who put the main blame on the airport company, said the owner’s requests in the first three years of the build had added 18 months of planning time.
The paper lists familiar ways extra work slips in. A request is small, so the team just does it. Someone asks a developer for a change “as long as you are in there.” Or the team adds features it finds useful or interesting, which the authors call gold plating. They also warn about the opposite: a change process so rigid that good changes get turned down.
The kitchen
Imagine a kitchen renovation with a fixed price and a date. In week two, since the wall is open anyway, the owners ask to move a socket. In week three, since the floor is up, they add underfloor heating. Each request sounds like an afternoon’s work. But the electrician has to come back, the tiler waits for the heating, and the plasterer loses a slot. The date on the contract never moved, so the last weeks get squeezed, and the finish may suffer.
A cheap fix is a one-page change log. Every addition is written down with what it will cost and how many days it will add, and the person paying says yes or no. Plenty still get a yes. They are just chosen with the price showing.
When the agent widens the task
With AI agents, the extra work can come from the one doing the job. Anthropic’s guide to prompting Claude Opus 5 says the model “can also expand the scope of a task, adding steps that weren’t requested or applying its own judgment about what the task should be.” Its notes on two earlier models describe extra files, unneeded abstractions and flexibility “that wasn’t requested”: gold plating, in a project manager’s terms.
The suggested fix reads like a good contractor’s rule. Tell the model that if a better approach exists, it should “say so in a sentence and continue with the task as asked rather than quietly narrowing, widening, or transforming it.” The idea still gets raised, and the person who owns the task decides whether it goes in.
On the human side, write the task down before the agent starts, and make each “while you’re at it” a separate request that can be checked and priced on its own.
Berlin had a written rule against changes, reaffirmed in the minutes. What went missing, by its chairman’s account, was the question each change should have faced: is it worth the risk?