Automation · AI Agents · Case Study

Count the publications, not the uptime: what 'zero manual intervention' has to mean

Published · Updated

Three case studies end on the same result line: zero human hours. Absence of effort proves nothing by itself, so we specify unsupervised work as a countable output with a period attached.

Three of our case studies end with the same number in the results table. Manual intervention required: 0, on a four-market e-commerce platform. Client time spent on operations: 0 h, on a consulting site. Weekly content operations time: ~0 h, on a personal media platform. It is the easiest figure to write down and the hardest to substantiate, because 'nobody touched it' is a claim about an absence.

Absence is not evidence

You cannot demonstrate that a system ran unattended by pointing at the thing you did not do. Absence of effort reads identically whether the pipeline worked perfectly or stopped entirely, because in both cases your week is empty. The only honest proof that an unsupervised process did its job is the artefact it was supposed to produce, counted at the frequency it was supposed to produce it.

What we specified instead

On the Tatano Energy platform, the figure we carry is 8 SEO articles published every day, driven by search trends, across 4 country domains indexed separately and 7 languages served. That is not a description of intent. It is a countable quantity with a period attached, which means anyone can check it from outside, without access to a dashboard, a log or an internal status page.

The personal platform states its cadence differently: articles published autonomously, daily, with weekly content operations at roughly zero hours. The structure is the same. One side of the sentence is the output, the other is the human cost, and both are observable without being inside the system. If the daily article is not there, the claim fails for that day, whatever anything internal happened to be reporting about itself.

A cadence is falsifiable, uptime is not

The practical difference is who is able to disprove it. 'The agent is running' is a statement about an internal condition, and internal conditions are reported by the same system that may be wrong about them. Eight articles a day across 7 languages is a statement about the world. Count the pages. Count the languages. The number is either there or it is not, and the shortfall is the incident.

Not everything is a cadence

Some of the results we publish are states rather than rates. Abbys Consult launched with time to first byte at < 200 ms and 0 technical SEO issues. Those are conditions measured at a point in time, and they can decay quietly without any publication going missing. The GEO audit score of 66→90+ on the personal platform is the same kind of figure: a measurement that was taken, not a stream that is maintained.

So the counting rule splits in two. Anything the system produces on a schedule gets a cadence you can count, and a missing item is the alarm. Anything the system merely is (response time, security headers, search signals, audit score) has to be re-measured deliberately, because no absent artefact will ever announce that it slipped. Two kinds of property, two different shapes of failure.

Where redundancy fits

The personal platform runs 3 model fallback levels. Redundancy of that sort exists to protect the cadence, not to replace the check on it. Fallback means a bad day for one model does not become a missing article. It does not establish that the article arrived, and the only thing that establishes arrival is the daily count. Build the resilience, then still count the output.

What to put in the brief

If you are specifying something that runs without supervision, write the deliverable and its period into the brief before you write anything about reliability. Eight a day. One a day. Then decide separately which point-in-time properties you will re-measure, and when. 'Zero manual intervention' is a reasonable thing to promise, but it is only credible as the by-product of a number somebody else can count.

Sources

Neurolinks case study — Four markets, one codebase — https://neurolinks.be/work/tatano-energy

Neurolinks case study — A personal brand that publishes itself — https://neurolinks.be/work/matthieu-pesesse-media

Neurolinks case study — A consulting firm, dressed for trust — https://neurolinks.be/work/abbys-consult

Working on a project where these methods apply?