A measure of tasks, published in the open.

Every headline about AI and work is either doom or denial, and neither tells a person which parts of their own week are actually moving. The useful unit is not the job title. It is the task. A bookkeeper does not do “bookkeeping”; they reconcile supplier statements, chase invoices, prepare a VAT return and talk to a nervous client about cash flow. Those four things have very different futures, and only one of them is a spreadsheet.

So we score tasks, one at a time, against a published rubric, and we let people see the whole breakdown for their own job, including the parts that stay firmly human, which are the parts most coverage of this subject never mentions.

The second reason is narrower and more honest: nobody currently publishes a like-for-like US and UK comparison scored by one rubric on one date, with the prompt attached. The UK is a first-class market here, not a footnote translated from American data.

Collab365 Futureproof is built and paid for by Collab365, a small British company that has spent over a decade running training and communities for people who work with Microsoft and, more recently, AI tooling. It sits alongside our other open experiments (a deterministic paper-folding engine, an AI research garden, an app-market forecast) on the AI Lab projects page, and it follows the same house rules: ground truth before prompting, every claim leaves a trail, reality gets the deciding vote.

The story of building it belongs on collab365.com, not here: what broke, what the crosswalk cost us, which scores were absurd, and what that taught us about the rubric. This site owns the research; that site owns the story of making it. Neither repeats the other, so the two cannot drift into disagreeing about the method.

We built Spaces because we are worried too. We have two daughters, and we can already see what AI is doing to the kind of work they are heading into. We wanted something that helps people move: into a new career if that is what it takes, or into a stronger version of the job they already have.

So, plainly. Collab365 sells memberships to Collab365 Spaces: paid, guided communities where people work through real problems with AI alongside others doing the same job. Some pages on this site link to a Space when one genuinely matches the work a visitor is looking at. That is the business model, and it is the whole business model.

What that means in practice, and what you are entitled to hold us to:

• The data, the scores, the task-level reasoning, the method and the downloads are free and complete. Nothing on this site sits behind an email address.
• The method and data pages carry no offer, no call to action and nothing promotional beside a number. The only mention of what we sell on those pages is the plain disclosure that runs in the footer of every page on this site.
• No score is nudged in either direction. The prompt is published and the model is named, so a thumb on the scale is detectable by anyone with an API key, which is the strongest guarantee we can offer, and considerably stronger than a promise.
• Where no Space genuinely fits, no Space is shown. An empty slot is a legitimate outcome.

We are not claiming to be independent. We are claiming to be checkable, which is the more useful property.

If a score looks wrong, a figure looks wrong, or a job’s description does not resemble the job you actually do, tell us: hello@collab365.com. The most useful message names the release id, the occupation, and what you think is wrong. The release id is at the bottom of every page carrying figures.

Corrections are published, dated, in the changelog. Scores are never quietly hand-edited: a task is re-scored by the pipeline, or a whole release is retracted and replaced, and both leave a public trail. If you have re-run our prompt yourself and landed somewhere different, we would particularly like to see it.

• Tell you that you are going to lose your job. We cannot know that, and claiming it would be an invented certainty.
• Rank “the most doomed jobs”. It makes a good headline and a bad instrument.
• Hide the occupations we have not scored yet, or the ones where the source data is too thin.
• Put anything we sell next to a number.