BEFORE YOU PRESS PLAY
What is discussed
AskMonty AI editorial noteSource consulted:
Montemagno describes an experiment within Spiegamelo Facile, the startup he says is operated largely by AI agents. The headline concerns a total advertising allocation of €100,000 across several campaigns, rather than money already fully spent or profit already earned. At the time of recording, he estimates that around €20,000 has been used. He states from the outset that he supervises the work. The experiment asks whether someone without programming expertise can use agents for real business operations involving money and practical consequences. It is presented as an ongoing personal trial rather than a completed, independently measured performance study. He had previously relied on specialist professionals and agencies to carry out this work. The starting point is an agent connection to Meta's advertising system. Montemagno says he began with €100–€200 tests and watched the system create images, write copy, configure advertisements and ad sets, and allocate budgets. He also describes changes to website conversion tracking. Additional advertising-related capabilities are intended to help the agents examine historical performance, acquisition costs and previous results. What particularly appeals to him is delegating an operational interface he has always found complicated and unpleasant to use himself. He sees value in being able to state the intended outcome while the assistant handles much of the configuration work. Part of the advertising promotes a free introductory event leading to a paid workshop. Montemagno mentions approximately 20,000 total registrations, while acknowledging that some come from his existing audience and organic content. He therefore does not establish an isolated return from advertising or provide a complete comparison with a human agency. His enthusiasm chiefly concerns the system's ability to carry out the work. He also says experienced advertising professionals could achieve more with the same tools, whereas beginners may fail to notice mistakes or poor settings. Access to automation does not mean that every user brings the same knowledge or judgement. The episode repeatedly returns to risk. Incorrect settings could spend money quickly or damage data and services to which an agent has access. Montemagno discusses backups, recovery planning and permission boundaries, and does not suggest that beginners should immediately adopt his own level of exposure. He also describes using browser control to change email-marketing automations, finding it more accurate than before but still imperfect. The practical distinction is deciding where an assistant may act and how closely its work should be controlled. More access makes the assistant useful in more situations, while also increasing the consequences when something goes wrong. His broader argument shifts from comparing models to the surrounding infrastructure. He uses “harness” to mean the method, context, tools and organisation that make an AI system effective. He wants a personal assistant able to use different models without becoming completely dependent on one supplier. He compares this with content creation: a durable method lets a creator adapt when social platforms change. Similarly, a coherent working system should make switching models easier. Control over data and accumulated context remains part of that goal, whether the system runs locally or in a dedicated remote environment whose arrangements the user can manage. Finally, Montemagno describes a change in his perception of personal limits. Tasks he had previously ruled out because he could not execute them become things he can attempt with agent assistance. A small team in which each person has assistants may substantially expand its output. People also need time to adjust their expectations and learn how to direct those capabilities. His conclusion is not that delegation removes oversight: opportunities, personal goals and tolerance for risk must be considered together when deciding how much responsibility to give the system.
Key points
- The headline refers to an advertising budget: €100,000 allocated across several campaigns with human supervision; the speaker estimates about €20,000 had been spent when recording.
- Agents handle operational work: creative assets, copy, ad sets, budgets and tracking configuration, according to his account.
- The example promotes a webinar and workshop: he mentions roughly 20,000 total registrations but acknowledges overlap between organic and paid traffic.
- More autonomy brings greater consequences: he discusses mistakes, permissions, backups and recovery, alongside improved browser use for email marketing.
- The system around the model matters: context, method and tools should allow suppliers to change without rebuilding everything.
- Perceived personal limits can change: assistants expand what individuals and small teams can attempt, while professional expertise remains valuable.
