How to have Claude run your company

Tech setup overview (Claude as “company OS”)

• A founder claims $250k/month revenue with zero employees, running the business entirely via Claude plus “cloud code.”
• Core architecture: business tools and data are connected to Claude, so Claude can:
  • generate content and assets,
  • read historical documentation (“second brain”),
  • query live analytics and other data sources,
  • execute automated updates/deployments,
  • run scheduled “agents” that communicate via channels like Telegram.

Product / tool stack & automation flow

• Claude + Cloud Code / terminal workflows
  • Everything is run from the command line (with notes about possible desktop app / cowork workflows).
• Model fallback
  • When the main system/model isn’t available, they use “JLM 5.2” as a cheaper alternative.
• Content + analytics connected to Claude
  • “Content stack” tools and analytics dashboards are linked so Claude can assess performance and produce insights.
• WhisperFlow
  • Used to speak faster (~150 words/min) instead of typing (~100 wpm).
• Post Hog
  • Claude accesses behavioral analytics such as scrolling and engagement.
  • Used for a CRO feedback loop (conversion benchmarks and improvement suggestions).
• GitHub repo + Next.js website
  • Landing pages are built as code in GitHub for easier duplication/versioning.
• Vercel
  • Used to deploy generated/updated site pages.
• Sentry
  • Error tracking for deployments.
• Cubic
  • An AI code review tool that checks Claude-generated code before going live.
• Playwright
  • Used for self-testing/inspection of the generated site behavior.
• Kit.com
  • Claude uses a Kit API key to create/manage forms, tags, and lead-magnet funnels.
  • Automation includes: creating forms, applying tags, connecting funnel steps, and routing subscribers into the right workflow.
  • Manual work is mostly limited to writing the nurture/email sequence content.
• Resend
  • Used for transactional emails to protect deliverability (described as specifically supporting Kit “checkpoint emails”).
• Atio (CRM)
  • Claude can create CRM records for leads and optionally enrich lead data.
  • Enables ICP-style targeting based on attributes discovered through signups.

Tutorial: generating an animated LinkedIn infographic + lead magnet funnel

Step-by-step described in the video

1. Infographic generation

   • Claude creates the infographic workflow from a prompted specification.
   • Under the hood: the infographic is an animated HTML page, then converted frame-by-frame into an animated GIF.
   • They note that LinkedIn GIFs can still work like images for reach/engagement.

2. Wireframe-first approach

   • Before producing the final design, Claude generates a wireframe (black-and-white layout).
   • Purpose: validate layout/spacing and readability for LinkedIn scrolling behavior.

3. Performance benchmarking

   • Claude scans an existing lead magnet landing page to extract conversion numbers.
   • The new page’s target conversion is set relative to the benchmark (example cited: ~45% target).

4. Landing page generation

   • Claude builds a lead magnet landing page on a Next.js site.
   • It runs a local preview (“localhost”) before deployment.
   • Deployment is handled via Vercel, with additional safety checks using Sentry/Cubic/Playwright.

5. Kit.com funnel wiring

   • Claude creates the Kit form, tags subscribers (e.g., “LinkedIn lead magnet”), and connects signup to the correct funnel.
   • Email sequence automation is supported by Kit; only the email copy is partially manual.

CRO / review loop (analytics-driven iteration)

• Claude is configured to ask clarifying questions (story angle, funnel structure, content realism, static vs animated).
• Claude also performs data-driven iteration:
  • Uses Post Hog signals to understand engagement and how users scroll/stop.
  • After generating the page, Claude can be asked for CRO recommendations (example: “three conversion rate optimization tweaks”).
• They emphasize that multiple CTA opportunities and form placement are part of the optimized template.

“Second brain” knowledge + permissions

• The founder maintains an “LLM wiki” style knowledge base:
  • 2–3k+ markdown files in a main “second brain” folder.
  • Historical context is stored in files; live data is connected to tools.
• Claude can query live systems (analytics, SEO, Lighthouse/performance, etc.) because permissions are configured safely, so it can act without constant manual approval.

Scheduled agents + omnichannel outreach

• Additional Claude-based agents handle operational tasks:
  • A “Jamie / James with AI” agent sends daily Telegram updates including pipeline status and estimated revenue timing/probabilities.
  • It can “see” pipeline context via tools like Atio, plus the user’s to-do list and other connected systems.
• Versel Eve framework:
  • Highlighted as the way to package/run agents.
  • “One command” in a folder turns that folder into a cloud agent (includes config/agent logic and channels).
  • Example: a “health agent” that pulls data from wearables (Garmin/TrainingPeaks/Hevy) to deliver daily and session-specific summaries.

Example: health/fitness agent pipeline

• Wearable integrations:
  • Garmin, TrainingPeaks, Hevy, plus daily metric nudges and weekly retros.
• Claude coaching-like insights:
  • Interprets training outputs (HR, pace, TSS/TSS variance, etc.).
  • Adds interpretive context (e.g., sleep/stress relationship).
  • Adjusts tone based on data quality (e.g., “pinch of salt” when sleep deprivation is evident).
• Data handling note:
  • The founder mentions they had to structure/condense health data into CSV/dashboard format because Claude struggled with raw volume.

Main speakers/sources

• James (interviewer/host)
• Andrew / the founder (main demonstrator; described as running the AI-driven company and Claude/automation workflow)
• Zapier (mentioned as “Presented by Zapier” / sponsor)
• Versel Eve (referenced during scheduled agents / cloud agent packaging)
• Kit.com, Post Hog, Sentry, Cubic, Playwright, Atio, Resend, WhisperFlow (tools forming the described system)