By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

The Calum Johnson Show

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

Matt Ebso uses Claude to run his business & life. And this year, his Claude agent saved him $20,000. In this episode, Matt builds AI agents live and shows you the fastest way to use Claude to save time in your business and your life.

Calum JohnsonhostMatt Ebsoguest

Nov 2, 20261h 21mWatch on YouTube ↗

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

CHAPTERS

Jump to publisher timestamps

Chapter summaries

Treating Claude like your operating system and first AI hire
Calum opens by asking what viewers will gain, and Matt frames Claude as an intermediary between you and your computer. Matt describes shifting from keyboard-and-mouse work to speaking into Claude as the primary interface.
- •Matt says the goal is a high-level understanding of using Claude like an intermediary with your computer
- •He frames AI tools as a “first hire” that can take on tasks in business and personal life
- •Calum asks what Matt’s desk workflow looks like and what value it creates
- •Matt describes using Claude’s microphone/transcription for stream-of-consciousness input
How Matt’s agent saved money by working from deep personal context
Matt explains that his agent has broad access to his files and email, allowing it to proactively find opportunities. He shares examples where the agent surfaced concrete savings and prepared drafts to send to his accountant.
- •Matt says his agent has access to email, projects, bills, and accountant conversations
- •He describes doing an hour-long brain dump to define outcomes and ask the agent to dig deep
- •Matt recounts the agent identifying a tax form that could save him money and drafting an email to his accountant
- •He shares a second example where the agent flagged a high electricity rate and suggested steps to reduce it
- •Matt emphasizes managing an AI like an employee: define the job, give context, set a schedule, review results
Time freedom philosophy: delegate, automate, eliminate
Calum asks about the influence of The 4-Hour Workweek, and Matt explains the motivation behind optimizing for time freedom. He outlines a framework for reclaiming time by delegating, automating, and eliminating tasks.
- •Matt describes “new wealth” as time freedom and optionality rather than just money
- •He lays out a framework: delegate what others can do, automate what can be automated, eliminate what can be removed entirely
- •Matt notes you shouldn’t delegate what can be automated because automation avoids ongoing human effort
- •Calum connects the idea to spending less time in “the muck” and more time in unique ability
Live Cowork demo setup: automating recurring haircut bookings
Matt begins a live demonstration in Claude Cowork by giving an agent a detailed prompt to find and book haircuts on a recurring schedule. The prompt includes decision rules, autonomy boundaries, and a request to turn the process into a scheduled routine.
- •Calum suggests a location for the demo; Matt sets it up as a fictional scenario
- •Matt prompts the agent to find a barber based on hair type, reviews, and easy online booking
- •He instructs the agent not to seek approval for low-risk decisions and to handle changes if a preferred option isn’t available
- •Matt asks the agent to define its process, then convert it into a recurring scheduled event
- •He explicitly labels the request as a demo to avoid confusing the agent’s prior context
Prompting for autonomy: preventing endless clarifying questions
Calum highlights key lines from the haircut prompt, and Matt explains why they matter. He compares agent behavior to a new hire who lacks a job description and will ask questions unless constraints and permissions are clear.
- •Calum points out the instruction to not come back for approval to support autonomy
- •Matt explains agents default to caution and will ask questions unless given clear permission and constraints
- •He describes Claude as “super genius with amnesia” that must re-parse context each turn
- •Matt recommends thinking like a manager: provide what a new hire needs to act without constant clarification
The 3-part prompting structure: context, task, format (Barcelona housing workflow)
Matt shares a detailed example of using Claude to tackle the Barcelona apartment hunt by turning it into a repeatable research workflow. He explains his three-part structure—context, task, and format—and shows how feedback loops improve results over time.
- •Matt describes the difficulty of apartment search competition and the need to respond quickly
- •He explains the framework: brain dump as context, then define the task, then define the output format
- •The output format is a spreadsheet with listing details, an AI-generated fit score, and human feedback columns
- •Matt describes a feedback loop where the agent reviews prior feedback, records learned preferences, and improves subsequent runs
- •He contrasts this with the time he would have spent building manual spreadsheet systems and research processes
Business workflows: research, SEO, and monitoring competitors with routines
Calum asks for business examples, and Matt explains how agents can handle operational and marketing tasks around the core work. He describes using Claude for research, SEO content generation, and maintaining accuracy by monitoring sites and repositories.
- •Matt uses a “goat farm” analogy to show agents can handle digital/admin tasks even if they can’t do physical work
- •He says he uses Claude heavily for research in his software tutorial work
- •He describes using routines for SEO and content ideas to surface a platform to more people
- •Matt says the agent can monitor competitor sites, check for outdated information, and draft articles
- •He notes the agent can access a code repository and combine app context with web research to produce publishable drafts
Exploratory pass → scheduled routine: building a guest-research engine in Cowork
Matt demonstrates an exploratory pass where Claude designs a repeatable framework to research and propose podcast guests. Calum and Matt add criteria, and Matt shows how Cowork turns the work into a structured plan with connectors and scheduled delivery.
- •Matt proposes building a weekly system that recommends a guest based on what has worked historically
- •Calum adds criteria like prior podcast experience and ability to handle long-form conversations
- •Matt emphasizes depth: find uncovered opportunities and low-hanging fruit the team might miss
- •Claude asks clarifying questions (voice/tone, guest constraints, timing, analytics access) to finalize the workflow
- •Matt highlights Cowork’s to-do list view and explains connectors as the toolset for an AI “employee”
Connectors and product modes: Chat vs Cowork vs Code (and model selection)
Matt explains how different Claude modes map to different kinds of work, and how connectors enable execution across tools. He also discusses choosing stronger models for exploratory work and lighter models for routine runs.
- •Matt says Chat behaves like a consultant answering questions, while Cowork acts more like a project manager/assistant
- •He explains Cowork can work in project folders and leave durable work outputs on your system
- •He reframes Claude Code as “your developer,” not a tool only for developers, and mentions built-in commands like looping checks
- •Matt discusses using higher-powered models for deep exploratory passes and lower-powered ones for repeatable routines
- •Calum frames the mindset shift: Claude becomes the interface that can pull from tools like email, calendars, and financial software
Building no-code-feeling apps with Claude Code: meal planning and grocery automation
Matt shows a hyper-specific meal-planning tool he built with Claude Code to reduce the weekly burden of planning meals under dietary constraints. He explains how it runs like an app on his phone, generates a weekly plan, builds shopping lists, and improves through feedback.
- •Matt explains the real problem is everything around cooking: finding recipes, adapting to constraints, and assembling shopping lists
- •He describes brain dumping constraints and preferences so the system can make rule-based decisions
- •He shows a progressive web app (PWA) experience with an icon and phone-like interface
- •The app generates a weekly meal plan, allows swapping meals with reasons, and learns from ratings and feedback
- •It prepares a shopping list and preloads a grocery cart, reducing the weekly planning and purchasing overhead
When to use Claude Code vs Cowork: assistant vs developer framework
Calum asks when someone is ready for Claude Code, and Matt answers with a role-based distinction rather than a readiness threshold. He explains Code is best when you want something persistent that can run without continuously consuming AI usage.
- •Matt suggests choosing the tool based on the role you’d hire: assistant → Cowork, developer → Code
- •He says Code fits when you wish an app existed, need a technical system, or want a process that runs without ongoing AI interaction
- •Matt emphasizes persistence: things built in Code can keep working even if you stop paying for Claude later
- •He notes Cowork and Code can overlap, but Code is ideal for deeper technical builds and durable automation
Demo results and closing playbook: feedback loops, skills, and first action step
Matt reviews the guest-research output Claude produced and shows how to refine it with feedback and turn the process into a reusable “skill.” They close with a simple starting workflow: voice-dump pain points, let Claude propose what to automate first, then schedule and iterate.
- •Matt walks through the generated guest research report and discusses how it proposes formats and candidate scoring
- •Calum reacts to suggestions, and Matt frames that reaction as essential feedback for the next run
- •Matt recommends reviewing each scheduled run, providing feedback in the same conversation, and storing it durably for future context
- •He explains converting an exploratory process into a repeatable skill/recipe for scheduled execution
- •Matt’s recommended first step: identify pain points, use the microphone to brain dump context, ask what to build first to save an hour or two per week, then let Claude run and iterate

Publisher timestamps

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.