One year of writing this: what we got right and wrong
The first post here appeared on 13 October 2024 and asked where Swiss SMEs actually stood. A year later, the honest scorecard.
This blog began on 13 October 2024 with a post called « Where Swiss SMEs actually stand with AI ». Twelve months and a great many model releases later, it is worth marking the date by checking what held up.
What we got right
That model choice would matter less than it appeared to. In October 2024 the assumption was that picking the wrong provider was a strategic error. It turned out to be a configuration decision, and by August every major supplier offered both hosted and downloadable options.
That the EU AI Act would not land on Swiss SMEs. This blog said so in October 2024, in January, in March, in April, in July and twice in August. The 2 August deadline passed and nothing happened, which is exactly what was forecast and the opposite of what most coverage implied.
That the real problems are unglamorous. Staff who do not use the tool. Nobody owning the decision. Accounts belonging to people who left. A year of writing has not turned up a more common failure than these.
What we got wrong, or at least understated
Version stability. In October 2024 this blog treated model updates as a minor operational matter. August 2025 showed it is a supplier-risk question, and the January posts should have said so.
Open models. The prediction in January was that downloadable models would remain a niche for firms with absolute confidentiality needs. That is still true for Swiss SMEs, but the speed at which the niche became the default for developers was underestimated by roughly a year.
What we would say differently to a reader starting today
Less about the technology, sooner. The most useful posts of the past year were the ones that were barely about AI: write the process down, name a person, ask five colleagues what happened, review the account list.
Thank you for reading. The industry will produce another two hundred announcements before next October, and roughly four of them will matter to a company of thirty people. Sorting those four out remains the job.