video1.0<iframe src="https://www.loom.com/embed/b89e822de67049dfb6494c8af0d5227e" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/b89e822de67049dfb6494c8af0d5227e-c6f0f2b2a0040898.jpg9338.31Host Samson presents a workshop on building and using AI assistants for business automation, focusing on practical examples, AI agent architecture, and tool integrations. Participants were polled on AI usage levels and discussed challenges like trust, security, and usability; Samson shared multiple real client cases showing automated accounting, IG/LINE-based registration flows, and a custom WordPress CMS integrated with AI. Next steps implied: explore AI agent tooling (N8N, connectors), choose AI models by task, and prototype persona-driven agents to automate repetitive business processes. ### Workshop topic and objectives 1:20 - Host announces the session topic: creating an AI assistant and career-focused support for people wanting to start businesses. - Goals include helping attendees discover career inspirations and reduce operational workload using AI. - Dates referenced for related events: May 21 and June 24; a six-week campaign launching June 30 mentioned. - Format: storytelling style with audience interaction, no formal PPT; practical steps offered on request. ### Audience poll on ai relationship and usage levels 3:50 - Host asks attendees to indicate frequency/familiarity with AI using numeric responses (1–4) mapping to heavy daily integration, tool-based use, consultative use, or rare use. - Defines levels: 1) full AI-integrated business with AI teams; 2) multi-tool users (ShareGPT, Gemini, etc.) who generate content and connect tools; 3) consultative single-AI users; 4) rare/novice users. - Host explains examples of automations and integrations (TrackGPT connecting Gmail, Google Assistant) and asks users to share what they do with AI and challenges faced. ### Interactive exercise: goals and challenges with ai 11:31 - Host gives attendees five minutes to answer two questions: what they want to learn about AI today and what challenges they face when using AI. - Host shares a simple chart and requests participants to post goals and problems in chat; music played during exercise and confirmation checks performed. - Intention: gather concrete user needs to shape the storytelling and practical guidance that follows. ### Security and trust concerns with ai 19:24 - Host reviews attendee comments and probes reasons for distrust of AI, asking whether concerns focus on data leakage to hackers or AI causing damage (e.g., deleting files). - Attendees express fear of AI 'doing damage' and security breaches; host acknowledges personal use of enterprise-grade providers (Anthropic/Cloud) and trust trade-offs. - Host describes provider safety prompts and confirmation dialogs (e.g., Cloud asking repeated confirmations when accessing repositories) and suggests isolating AI access on a dedicated 'clean' machine as a mitigation. ### Case study: automating accounting and customer flows 26:37 - Host describes a client engagement where AI automations were built to manage company accounting and reconcile payments using Google Sheets/Notion and bank connectors. - Implemented features: automatic reconciliation when orders arrive, LINE group-based bookkeeping inputs captured by an AI agent that fills reports, and automated confirmation flows (e.g., reply codes to confirm payments). - Host highlights value: reduces repetitive tasks, unblocks founders who do not like numbers, and frees staff from manual checks; notes bank connectors provide read-only access to transaction data (no direct transfers). - Host also mentions an IG registration helper demo and automated course enrolment flows triggered by payment confirmation within LINE/Telegram. ### Building ai agents: tools, architecture, and n8n workflows 32:41 - Host explains the concept of AI agents and layers of automation: conversational assistants, agents that use connectors/tools, and programmatic/skill-based execution. - Describes using N8N as a visual workflow/orchestration platform to chain AI agents, trigger from messaging apps (Telegram/LINE), and integrate with Google Drive, Calendar, forms, and other tools. - Emphasis on creating AI 'characters' by setting system messages/personality, uploading a knowledge bank, and providing work manuals so agents follow brand tone and procedures. - Host warns about fragility and the need for careful command design; shared that early setups required debugging and careful prompt/command structure. ### Comparing ai model personalities and capabilities 59:03 - Host compares multiple AI providers (ChatGPT, Anthropic/Cloud, Gemini, Grok/XAI) and maps perceived personalities and strengths to tasks. - Observations: ChatGPT provides high emotional/communicative value but can be flippant; Anthropic/Cloud is meticulous and strong at execution and connectors; Gemini excels at multimodal understanding (videos) and structured teaching; Grok is creative and connected to social feed (X/A), good for trend/market insight. - Host recommends selecting models by task (execution, creativity, system architecture, multimedia understanding) and switching models when outputs are unsatisfactory. - Notes connectors and tool integrations vary by provider; some providers offer Codex/Codecs for programmatic control and stable automation. ### Vision: ai team as company assistant and productized services 01:16:19 - Host shares ambition to assemble an AI team that can manage business operations, letting the host focus on content, appearances, and strategy. - Describes building a custom CMS/WordPress-based system integrated with AI for reservations, finance, product pages, automated emails, community broadcasts, and customer lists. - Notes client demand for productized all-in-one platforms (e.g., GHL) and the aim to offer AI-enabled automation services so clients can delegate operations and focus on core offerings. - Host states practical limits: humans still make decisions; AI can automate repetitive tasks and partially execute decisions, but full replacement not claimed; encourages prototyping persona-driven agents and choosing appropriate tools for tasks. ### Project overview and motivation 01:30:56 - Samson built a custom CMS integrated with AI to automate marketing, backend operations, and product workflows. - The CMS connects to cloud services and third-party tools (Stripe, Telegram, LINE, Notion, messaging, community software) to automate end-to-end processes. - Goal: let AI handle operational tasks so Samson can focus on coaching and client work instead of manual backend maintenance. ### Complexity, maintenance burden, and feature bloat 01:31:55 - Samson reflects that full-featured platforms require many people to maintain numerous small functions (buttons, notifications, modules). - Each feature requires lifelong maintenance, debugging, and updates, which becomes a major overhead for non-engineers. - Realization: building a huge feature set may not align with personal values or lead to sustainable operations. ### Experimentation, multitasking mindset, and learning approach 01:35:02 - Samson adopted a playful, experiment-first mindset: try many approaches (thousands of iterations on GitHub, hundreds of workflows) to increase chance of success. - Emphasized combining intuition and logic, treating AI learning as a game rather than anxiety-driven training. - Persistence alone is not the metric; iterating widely and having fun increases learning opportunities. ### Health crisis and reevaluation of priorities 01:40:57 - Samson describes contracting a bacterial infection abroad (Cotyledon), leading to severe symptoms, hospitalization, kidney failure indicators, and a month of immobility. - The illness prompted rethinking of career values and work practices, emphasizing health and simplifying workload. - Recovery progress noted (about 70% better) and recognition that earlier disregard for body signals contributed to the situation. ### Shift to ‘do less, get more’ and product reassessment 01:45:28 - Post-illness, Samson prioritized simplifying his software and services: removed unnecessary features and focused on core offerings. - Found that many client-requested automated functions were unused or abandoned; clients sometimes needed clarity rather than more automation. - Decision to return emphasis to coaching/consulting and to make AI tools support core human interactions rather than replace them. ### Demonstration of ai assistant workflows and personal use 01:50:44 - Samson demonstrated using a Mac Mini hosting an AI assistant and showed examples of composing and sending marketing emails via voice and Telegram while resting. - AI drafts emails, selects recipients (e.g., those who opened previous emails), prepares Notion pages, and reports metrics on demand. - Samson prefers minimal manual backend interaction; AI handles segmentation, timing, and content generation with human review when needed. ### Customer service automation and guardrails 02:00:21 - Samson uses AI to draft customer service responses and surface relevant registration or class information, while human staff review and send replies. - The system flags when a student appears already registered and suggests confirmation steps rather than redundant sign-ups. - Emphasis on maintaining human oversight to avoid random or incorrect AI replies; automation is used to speed response, not fully replace humans until volume demands it. ### Voice assistant features and mobile use cases 02:12:24 - Samson built a voice-based assistant (Eva + manager) to interact via phone for hands-free tasks while traveling or driving. - Use cases: ask for daily radar reports, schedule tasks, instruct AI to notify team members, and receive spoken summaries when screen use is impractical. - Feature described as partly playful and not strictly necessary, but useful for specific scenarios where hands-free interaction is safer or more convenient. ### Principles for others: relevance, simplification, and training ai 02:20:55 - Advice: treat learning AI as a game; prioritize relevance to one’s career and core services instead of chasing comprehensive toolkits. - Recommend starting with minimal features that support core value (e.g., coaching one-on-one) and training AI to internalize brand voice and processes. - Offerings and next steps: one-on-one collaborations to design AI-led workflows, a free sharing session on June 24, and a structured class beginning June 30 for managers to organize business with AI. ### Closing summary and calls to action 02:27:48 - Samson reiterates two main points: don’t start AI from anxiety—approach it as play—and favor 'less but better' over feature expansion. - Encourages attendees to join upcoming sessions and one-on-one coaching for tailored AI automation support; mentions discounts for previous course participants. - Final asks: attendees reflect on most valuable takeaways and consider whether to pursue simplified AI solutions aligned with career goals.