All posts / ai-at-work

AI that actually works

Agents doing real work: automation, orchestration, governance. Receipts included.

4 October 2026 · AI that actually works · 3 min read

Ellissi· AI

How to take apart a tech prophecy in five questions

Elon Musk says money won't matter by 2036. Ellissi's method to take apart any tech prophecy: five questions — audience, evidence, inevitability, frame, and whether the conclusion exceeds the premises.

Read →

23 September 2026 · AI that actually works · 4 min read

Ellissi· AI

The lemon that bought a Schweppes (and other AI dinners)

Settavola plans the week's menu and prepares the shopping for a single family. The AI proposes, the family approves, the system doesn't pay. In between: a Pepsi instead of sugar and a constraint that lived only in the prompt.

Read →

18 September 2026 · AI that actually works · 3 min read

Ellissi· AI

All roads lead to AI, might as well join the club. No

The determinism of those who build AI and remove themselves from the picture to shed responsibility, versus the practice of fences, kill switches and human signatures. AI is not a force of nature.

Read →

15 August 2026 · AI that actually works · 3 min read

Ellissi· AI

A 2012 text where I'd already written today's doctrine

Lead the Future, Ambrosetti Forum 2012: 'you have to fail to understand how it works', 'iterative development on high-frequency feedback', the re-startup process. Fourteen years before the agents, it was all there.

Read →

27 July 2026 · AI that actually works · 3 min read

Ellissi· AI

My Constitution reached version 8 in thirty days

Post-mortem of a governance runaway: how a one-person company ended up with 483 lines of fundamental law, zero customers, and a self-imposed constitutional moratorium.

Read →

15 July 2026 · AI that actually works · 3 min read

Ellissi· AI

The kill switch: why each of my companies can die within an hour

The constitutional rule that helps me sleep: every company that goes to market must be stoppable in under 60 minutes, and the switch must be tested on day one. It's engineering — but above all, it's trust, provided the test is real.

Read →

14 July 2026 · AI that actually works · 3 min read

Ellissi· AI

I hired ten employees who never sleep (and don't exist)

Welcome to the Gargency chronicles: a holding with one human CEO, a 110-line Constitution and a workforce of AI agents doing real work. For better and — with better storytelling — for worse.

Read →