Revised from my Medium article “The catastrophe I keep waiting for”, published on 28 April 2026. The project experiences below are those reported at the time.
Completing work faster releases time. What that time is worth depends on how a business uses it. It can make room for a previously unaffordable project, improve the attention given to existing work or support additional orders. I have seen sharply different outcomes, and they changed what I expect from an AI business case. An estimate of hours saved is incomplete without a decision about where those hours will go.
A project becomes viable
I led the technology and architecture of a customer, sales and marketing data platform at a European midsize company. It entered productive use four weeks after kickoff, designed for a user base of around 60 people. The system used an open-source stack on the customer's infrastructure. AI assisted development, while the sensitive-data paths at runtime operated without a language model.
The earlier estimate assumed a small consulting team and an internal lead working for six months. That put the programme beyond both the budget and the timetable the customer could accept. It had spent about five years among the projects that would be commissioned when more money became available. The resulting work was therefore work that had not previously been done. This was one project, not a controlled comparison.
This distinction matters when assessing the economics. Faster implementation can make a new capability affordable even when there was no existing position or contract to replace. It gives a practical reason to reconsider deferred projects using a fresh estimate of their costs and requirements.
What one project says about the economy
Labour-market analyses distinguish between different effects. Goldman Sachs's 2023 discussion explicitly considers partial automation, new occupations and additional demand. Its widely quoted figure of 300 million full-time job equivalents concerns work potentially exposed to automation. It is not a forecast of 300 million redundancies. The IMF's 2024 analysis also considers both the substitution of human work and the ways AI can complement it. Goldman Sachs, 2023, IMF, 2024
My example leaves open how often AI makes previously unaffordable projects viable and how much that contributes to the wider economy. For an individual company, a concrete decision remains. Management must decide which work to commission, what standards it must meet and how any released capacity will be used.
The omission in my own business case
At a German technology-services firm, events took a different course. Proposal creation and integration development could be automated, and the first quarter's productivity figures initially matched expectations. A few months later, management asked sales to expand the pipeline on the assumption that automation would absorb the additional delivery work. The business case had not assigned the saved hours to a specific purpose. They became additional project volume.
Experienced consultants and engineers then had to divide their attention across several engagements. I saw some of the strongest people become overwhelmed within months as the quality of the work visibly declined. Management owned the decision to pursue more volume. I also owned an omission, because I had prepared the proposal without specifying the destination of the saved time.
A more cautious productivity estimate alone would not correct that mistake. A calculation can capture a genuine time saving while overlooking the pressure created by subsequent commitments. In this case, the missing link was between technical efficiency and the allocation of work.
Making the allocation explicit
A credible business case should identify which activity takes less time and who can use the released capacity. Several spare minutes between appointments are not equivalent to an uninterrupted working day. Review, corrections and follow-up questions belong in the calculation alongside faster production. Only then can management judge whether another assignment, more attention to existing customers or a shorter delivery cycle is realistic.
A named person should own the intended use and review whether it happens. Relief from pressure calls for different observations from a growth target. The former requires attention to workload and available time. The latter also requires evidence that delivery and quality hold as volume increases. If the intended result does not appear, the allocation needs another decision. The Kupermann Decision Partner offers a way to record those assumptions, responsibilities and reasons to revisit a recommendation.
Employees' views belong in that process. A time-consuming activity can also provide professional identity, customer contact or a chance to learn. FireScore starts with people's assessment of the recurring tasks they perform. That participation does not replace the financial case or the decision to be made. It can reveal which forms of relief people want and which consequences a process map has overlooked.
Keeping a route into experienced work
Junior development presents another practical question. Research, initial programming assignments and documentation can be opportunities to learn a discipline through practice. When those tasks are automated, work planning needs to provide new opportunities to learn. The open question is which assignments will help less experienced staff develop judgment and what support they will need to take responsibility.
My response is to make learning opportunities explicit in the allocation of work. Someone expected to review an answer needs more than access to a finished model response. Independent attempts, discussion of mistakes and feedback from experienced colleagues all deserve a place. The appropriate tasks and supervision will differ by profession. Buying more tools does not settle those choices.
In the midsize project, AI helped make a previously unaffordable system possible. In the other case, an unspecified destination for saved time contributed to excessive pressure. Those experiences support a deliberate allocation of both capacity and learning time. The business value of time saved depends on what people are able to do with it.