In February 1994, Denver let reporters watch a test of the automated baggage system at its new airport. According to The New York Times, bags “went flying, tumbling and bursting open, with some sliced in half,” and the press saw “smashed baggage and airborne underwear.”
The airport was supposed to have opened the previous October. The opening had already moved to December, then to March. Now the mayor moved it again. “I will not risk Denver’s reputation on an inadequately tested baggage system,” he said. In May the opening was postponed indefinitely. The city was losing $18 million to $19 million a month. The case has the rest.
The date is fixed, the scope is not
Fix the deadline, then decide what fits. When time runs short, cut scope, not the date.
That is a timebox. The DSDM agile framework defines one as “a fixed period of time, at the end of which an objective has been met,” typically two to four weeks. At the end, progress is judged by what is finished, not by how busy anyone was. A sprint is a timebox. So is a meeting that ends on the hour whether or not the agenda is done.
The framework is blunt about what usually happens instead. “In a traditional project, all requirements are treated as Must Have,” and “typically time (the end date) will slip if problems are encountered.” A timebox only works if something can give, so DSDM sorts the work before it starts: Must have, Should have, Could have, and Won’t have this time. It suggests Must haves take no more than 60% of the effort and keeps about 20% in Could haves as a buffer. Its test for a Must have: if the team said the night before launch that this one item couldn’t be delivered, would you call off the launch?
What Denver cut
For most of its history, Denver’s baggage plan ran the traditional way. The scope was fixed, an automated system for the entire airport, and the date gave. A consultant had warned the city in 1990 that such a system probably couldn’t be built and tested in time. The first two delays came from unfinished construction, but federal auditors found that the later ones were caused solely by the baggage system.
In July 1994 the city turned the plan around. It fixed a date, February 28, 1995. Read as a timebox, that decision made one thing a Must have: bags had to reach planes. Automation could wait. The city built a conventional tug-and-cart system as a backup, released the contractor from serving every concourse, and cut the required capacity from 65 bags per minute per line to 30. The airport opened on the chosen day, with the automated system handling only United’s bags on one concourse.
The fixed date didn’t make the automated system work. It forced a decision about what the airport couldn’t open without. The auditors later concluded that planning the backup from the start would have been cheaper.
Imagine a small bakery building an online ordering page and aiming to launch the week before Mother’s Day. If the owner treats the photo gallery and the loyalty scheme as just as essential as taking orders, the launch could slip past the busiest week of the spring.
Budgets for AI agents
AI agents need an end point too. Anthropic notes that an agent’s task “often terminates upon completion, but it’s also common to include stopping conditions (such as a maximum number of iterations) to maintain control.” In its multi-agent research system, it found that “agents struggle to judge appropriate effort for different tasks,” so it wrote budgets into the instructions: one agent with 3 to 10 tool calls for simple fact-finding, and more than 10 subagents for complex research.
A limit on its own only stops the work, much as a deadline on its own only ends the meeting. What made DSDM’s timebox useful was the sorting done before it started. Giving an agent a budget raises the same question Denver had to answer in 1994: when the budget runs out, what must already be done?
Denver’s airport opened on the day the city picked. It opened without most of the automated system it had waited for.