57 Minutes of AI training that will explode your business in 2026

What the session is trying to achieve

• The presenter’s stated goal is to help companies systematically adopt AI for business automation (2026 readiness).
• The path starts from what participants already use and moves toward:
  • Agentic automation (Claude/MCP)
  • An AI Operating System (AIOS) built with Claude Code

Participant “baseline” (inputs from a pre-survey)

The presenter reports that, prior to the session, attendees answered seven questions:

• AI usage frequency: ~100% use AI several times per week
• Primary use cases today: mostly research, thinking, writing content, and “doing business work without involvement”
• Most used tools: Claude, ChatGPT, Gemini
  • (with Codex/Zephyr as minorities)
• AI spend per month: mostly $50–$200, or $200–$1,000 (some > $1,000)
• Repetitive work hours that could be automated:
  • main group: 5–15 hours/week
  • smaller group: <5 hours/week
  • exception: 15–40 hours/week
• Process documentation: “some systematized, some not”
• Main blockers to adoption:
  • distrust (“I don’t trust the results”)
  • “I don’t know where to start”
  • it “broke” after trying
  • “no time to set it up”

Core business message / operating principle

• AI is positioned as something that can build and manage a system simultaneously.
• For founders, the opportunity is less about learning AI itself and more about delegating processes to AI via:
  • Correct prompting
  • Connecting AI to tools/files/processes
  • Packaging workflows into repeatable assets
• The presenter argues AI fails mainly due to bad inputs + missing context (“garbage in, garbage out”), not because AI “isn’t good.”

Frameworks / playbooks explicitly referenced

1) Claude “3 uses” framework (everyday work → product → artifacts)

Claude is described as having three main use areas:

• Everyday work
  • research/brainstorming
  • tool execution via MCP connectors
• Projects
  • a persistent “workspace” inside Claude with instructions + files + chat
  • improves via “project memory”
• Artifacts
  • creative/design/document generation inside Claude
  • e.g., presentations and doc-like outputs

2) AIOS / Claude Code OS architecture (context → data → decision)

To build an “AI Operating System,” the presenter outlines three layers:

• Context
  • who you are
  • business goals
  • team/roles
• Data
  • transcripts
  • sales/financial reports
  • emails/Slack
  • SOP libraries, etc.
• Decision
  • using past decisions as training examples
  • the presenter personally avoids fully delegating “live” business decisions

3) Automation prioritization: “High-Impact / Low-Effort” matrix

A selection matrix for what to automate:

• High impact + low effort → start here (“low-hanging fruit”)
• High impact + high effort → consider later
• Low impact + low effort → avoid
• Low impact + high effort → avoid

The presenter also advises identifying what consumes the most time personally (example: call analysis at ~20 hours/week).

4) Spark process for deploying automations (Systematize → Connect → Act → Repeat → Maintain)

A repeatable approach:

1. Systematize: map the process first (audit + process mapping)
2. Connect: ensure each tool in the workflow has an API / integration available
3. Act: get the AI to run the workflow end-to-end at least once
   • iterate until output is “perfect enough”
4. Repeat: package as a reusable “skill”
5. Maintain: deploy on schedule and monitor/iterate

Concrete examples / use cases (business execution)

Claude + MCP: connecting to operational tools

• Example tools mentioned for direct connection:
  • Notion, Typeform, Stripe, Shopify, Miro
  • (YouTube was mentioned, but uncertain)
• Example “agent acting” flow:
  • If Claude can access Miro, it can log in, perform tasks, and return results (without manual steps).
  • If the app lacks API access, fallback options are:
    • partial automation with a manual step in the middle, or
    • granting agent control over the computer (browser/desktop control)

Project example: client-specific content/script workflows

• “One client = one project” approach (Khalid example):
  • Project includes:
    • language mechanics
    • ICP
    • hook structures/formats
  • Outcome: generate scenarios per client and improve over time as conversations accumulate.

Quality control through frameworks + structured inputs

• The presenter treats AI like an “employee/screenwriter”:
  • Bad instructions/frameworks → bad output
• Therefore, quality at scale depends on building good frameworks (especially for sales/marketing structures).

Sales call analysis automation (Hardnox / presenter claim)

• Example workflow:
  • Upload sales call transcripts into a project
  • AI parses and produces feedback in the way the business wants
• Advanced version:
  • after each call:
    • transcribe → analyze → generate report in Google Docs → send to reps

Claude Code: skills + scheduling

The described automation lifecycle:

1. connect tools (API/MCP)
2. run the workflow once to validate inputs/outputs
3. package into a skill (SOP-like)
4. deploy to a scheduler/hosting platform (example: Modal)

Data centralization dashboards (main Claude Code use case)

• Emphasis: centralize metrics across many tools into a single “dashboard,” then ask questions on top of it.
• Example metric sets:
  • E-commerce: ROAS, creative created, daily sales
  • Deal/marketing ops: funds received, close rate, planned calls, qualified leads
  • Attribution: track which content drives clicks → leads → sales

Dashboard timeframe suggestion (practical GTM/ops choice)

• Recommended baseline:
  • connect apps first
  • pull last ~6 months of data to build the initial dashboard
  • then automate daily/weekly updates
• Note: historical data may be messy; whether it matters depends on whether history impacts future decisions.

Key terms mapped to “how automation works” (business-relevant)

• LLM: the “brain” model (ChatGPT/Claude as different model “types”)
• Agent: can perform actions (not just chat)
• API: how software communicates
• MCP: connector layer enabling an AI assistant (Claude) to access connected tools
• Protocol naming confusion acknowledged: GPT family/model versions may be mentioned, but the focus remains on Claude + Claude Code

Metrics / KPIs mentioned (explicit or implied)

No formal KPI targets were provided, but the session references dashboard contents and optimization metrics, including:

• Time metrics
  • repetitive work: 5–15 hours/week (survey)
  • example: ~20 hours/week on call analysis
• Sales performance KPIs
  • close rate
  • deal fit analysis (investor suitability)
  • qualified leads
• Marketing/e-commerce KPIs
  • ROAS
  • creative volume created
  • daily sales
  • attribution (content → clicks → leads → conversions)
• Ops KPIs
  • order fulfillment/delivery systems (described as the complex post-sale side)

Actionable recommendations (directly stated)

• Start by connecting AI to tools you already use (MCP/connected apps) rather than building complex custom integrations.
• Use the high-impact/low-effort matrix to pick the first automation.
• Build quality by:
  • creating strong process frameworks
  • providing correct context
  • using structured document formatting (see below)
• For legacy tools without APIs:
  • automate upstream + downstream, or
  • use computer-control fallback (with the user present)
• If hiring help:
  • learn basics first so you can validate outputs
  • the presenter is skeptical of agencies (“no context,” fast-changing field)
  • prefer staff or an in-house AI specialist once foundational understanding exists

Implementation details that affect execution (how-to specifics)

For Claude Projects: instructions formatting

• When adding documents into Claude projects, use Markdown headings (e.g., #, ##, ###) so the AI can structure content.
• Provide:
  • a role definition (“who the AI is”)
  • clear step-by-step tasks
  • input/output examples to match the organization’s tone

Generating system prompts

• Recommendation: use WhisperFlow to dictate prompts.
• Then generate a system prompt in Markdown for projects.

Claude Code “skills” packaging

• A skill is described as an SOP-like package:
  • instructions + prompt + steps
• Skills are created after running the workflow once end-to-end.

Scheduling automations

• To run daily automatically, deploy the packaged automation (example: Modal).
• Differentiation described:
  • Modal/Railway: hosting scheduled automations
  • Supabase/Obsidian: data storage/retrieval (when volume is large)

“When to stop doing it yourself / hire” (management guidance)

• Learn basics yourself; then delegate when:
  • maintenance/checking becomes time-consuming
  • you can’t justify ongoing monitoring overhead
• The presenter recounts learning by first outsourcing incorrectly (“client said it wasn’t what I asked”).
  • This led to self-learning Make.com/automation foundations so they could review quality.

High-level investing/market angle

• General claim: businesses risk falling behind if they don’t become AI-native within 3–5 years.
• The emphasis is on operational execution, not market analysis.

Presenters / sources

• Presenter: Michele
  • founder/provider of AI training
  • trained 5,000+ people in live sessions
  • worked with 50+ companies
• Referenced participants/cases: Khalid, Tristan, “Hardnox” (company mentioned)
• General tools referenced: Claude, Claude Code, ChatGPT, Codex, Grok Bot, and Claude connectors/MCP