The Man Who Mastered Claude: The Fastest Way To Build Your First AI Agent To Save Time (Live Demo)

Matt Ebso says he tries to do “stream of consciousness” voice input into Claude and then let the agent do the driving, with his role shifting to reviewing and giving feedback. He frames this as making Claude the interface through which work gets done, rather than manually hopping between apps.

Matt describes giving his agent access to email, bills, and accountant conversations, then asking it to “dig deep” for improvements; he says it surfaced an $18,000 tax-form savings and identified an electricity-rate issue, with an email draft prepared for his accountant. He presents this as the leverage that comes from context plus autonomy.

Matt attributes his automation mindset to the “new wealth” idea (time freedom) and a funnel of delegate → automate → eliminate. He argues you shouldn’t delegate what can be automated, and that many tasks should be eliminated entirely.

Matt recommends structuring prompts in three parts—context, task, and format—illustrated by his apartment-search workflow that outputs to a spreadsheet with fit scoring and a feedback loop. He emphasizes iterative improvement: run, review, add feedback, and store learned preferences.

In the haircut demo, Matt explicitly grants autonomy (“don’t come back for approval”), clarifies risk boundaries (no passwords or card payments), and asks Claude to first outline the process and then schedule it. He explains this reduces back-and-forth because the agent otherwise asks many clarifying questions.