Research

The people to follow.

Real founders running real businesses on Claude, Muse, and AI agents, organized by who we should follow most as we build our first thousand dollars, then ten, then a hundred, the honest way. No spammy schemes, no passive-income-in-a-box scams. Just operators, what they actually do day to day, and the numbers they claim.

Researched October 2, 2026. Every example names its source. Confidence labels say whether a number is vendor-reported, interview-claimed, or self-reported.

The patterns, before the people

Across every legitimate operator, the same practices show up. Learn these and you have the playbook:

Follow these people

The honest operators running one-person or tiny-team businesses on AI agents. Closest to what we are building at the Studio. Start here.

#01 · Follow these people

BlueVeta (Tomer Bar-Meir)

Real Stripe payments, small customer base, ~75% of per-customer work done by Claude. self-reported

Sole founder. Upload one product photo, get a full marketing campaign (studio shots, lifestyle scenes, video, platform-ready copy for 20 platforms) in minutes.

What the agents actually do: Every company function is its own persistent Claude Code session with scoped ownership of one domain, each in its own git worktree. Sessions message each other across domains and share docs as institutional memory. Hard rule: investigation is free, but nothing is built, deployed, or spent without coming to the founder first. The best operational detail of any solo case, the weakest revenue evidence.

Source: dev.to post

#02 · Follow these people

jamesai

Claims $1M+ lifetime revenue, 6 SaaS products, one person. self-reported

Solo founder running 6 SaaS products in parallel over 32 months: "just me, a laptop, and a Claude Code window open most of the day."

What the agents actually do: He open-sourced 25 Claude Code skills covering idea discovery, demand validation, product building, marketing, ads, autonomous operation, and cross-session memory. The skills are reviewable code, which is the most verifiable part of the claim: 6,500+ paying transactions, 10,000+ paying customers.

Source: show-me-the-money on GitHub

#03 · Follow these people

Evgenii Arsentev

Three $200 Claude subscriptions replace an estimated $38,500/month team. self-reported

Medical doctor and CEO running digital health businesses. No full-time developers, DevOps, designers, analysts, or product manager.

What the agents actually do: Each business line has its own scheduled agent shift on separate servers: deploys and releases, a custom-built CRM, hourly mailbox sorting, meeting scheduling, tech support, news publishing four times a day, open-source AI release scans three times a day, educational games built overnight with a separate agent checking screenshots. In one business line, an agent reviews call-center transcripts daily and writes the human supervisor's task list. Guardrails: every workflow has a named human owner, nothing ships without passing a check, agents get access for a task not a standing role, and the former CTO stays on a small retainer as a safety net.

Source: dev.to post

Steal these playbooks

Specific tactics worth copying: frameworks, pricing, campaign systems, and small honest wins.

#04 · Steal these playbooks

Nate Herk

Scaled an AI automation agency past $100K/month, then sold it. self-reported

Ran an AI automation agency for small businesses; now teaches the "AIOS" (AI Operating System) method to a large community.

What the agents actually do: Claude Code in VS Code as the layer between operator and computer, built on four C's. Context: local markdown files. Connections: APIs, CLIs, and MCPs into ClickUp, Google Workspace, Slack, Stripe. Capabilities: custom skill SOPs. Cadence: scheduled routines with a daily loop. Client playbook: niche down to one industry and one problem, value-based pricing, client owns API keys and infrastructure.

Sources: Calum Johnson interview, course transcript on GitHub

#05 · Steal these playbooks

Frey Chu

Builds niche directories for about $273/day claimed, $5 to 15 in API per build. self-reported

Builds automated niche online directories: dementia care facilities, ketamine clinics, immigration lawyers.

What the agents actually do: Claude Code builds each directory site in about 4 days for under $250 (Claude API, Cloudflare Pages hosting, Supabase free tier), using free public data sources and scrapers. Early stage: the $273/day is a podcast episode title, and portfolio projections ($5k to 25k/month across 10 sites) are the author's, not verified revenue.

Source: methodology repo on GitHub

#06 · Steal these playbooks

Jonathan Courtney

Claims a $450K webinar campaign built with Claude. self-reported

Founder and podcast host running marketing campaigns and webinars. His principle: use AI to run more campaigns (abundance and scale), not to cut headcount.

What the agents actually do: A four-step "Promoter Blueprint" (traffic, holding pattern, selling event, conversion) executed with Claude Projects plus Claude Code, going from research to a shipped marketing asset in under an hour. He warns AI becomes a "procrastination machine" when used to optimize systems with zero customers.

Source: YouTube interview

#07 · Steal these playbooks

KANE agency

Proposal pipeline compressed from weeks to minutes. vendor-reported

Creative and growth agency in Princeton, New Jersey.

What the agents actually do: A Claude pipeline scores prospect fit from meeting notes, prices from past projects, drafts the proposal, and gets the statement of work out for signature within minutes of a yes. Clients recognize the speed difference.

Source: Anthropic via Unite.AI

#08 · Steal these playbooks

Bambi Baby

$60,000 in sales tied to a Claude-built tool in four days. vendor-reported

Family-owned baby stroller and car seat retailer. The COO is not a developer.

What the agents actually do: He used Claude to build a scan-to-wishlist lead capture tool and the follow-up automation behind it. Two weeks into a two-store pilot, 18% of online revenue was newly attributed to in-person visits.

Source: Anthropic via Unite.AI

#09 · Steal these playbooks

HireEffect

Found 16 transactions behind a $13,000 discrepancy in 2.5 years of PayPal data. vendor-reported

Dallas bookkeeping, payroll, and HR firm.

What the agents actually do: Claude isolated the exact 16 transactions behind the discrepancy, buried in tens of thousands of transactions. A governed reporting dashboard across five systems turned a two-hour monthly task into ten seconds.

Source: Anthropic via Unite.AI

#10 · Steal these playbooks

Mothership Coffee Roasters

In-store margins rose to 22%, credited in part to Claude tooling. vendor-reported

Las Vegas coffee company with six cafes plus wholesale and catering.

What the agents actually do: One consolidated Claude view for scheduling, ordering, and inventory reporting across all locations. Tighter operations, credited by the founder in part for the margin lift.

Source: Anthropic via Unite.AI

Admire from afar

Impressive, but venture-backed teams with headcount. The agent patterns transfer; the scale does not, not yet.

#11 · Admire from afar

EvenUp

Firms using it resolved 200,000+ cases and secured $10B+ in damages. vendor-reported

AI platform for personal injury law firms. Claude models do the reading, extraction, and reasoning across case documents, with EvenUp's proprietary injury data and automated workflows on top.

What the agents actually do: Read and reason across case files, draft demand documents, pull past medical expense totals, and assemble settlement packages with cited evidence. Drafting went from 8 to 15 hours of skilled work per case to minutes. Settlement offers rose 300% for firms presenting the full damages picture; one practice cleared a 45-day backlog.

Source: Anthropic customer story, claude.com/customers/evenup

#12 · Admire from afar

Lovable (CEO Anton Osika)

$100M+ ARR in under a year, $1.8B valuation. founder-stated

Swedish startup that lets non-coders build working apps by describing ideas. The product runs on foundation models including Claude, and shipped an agentic mode that reads files, debugs, and pulls web results.

What the agents actually do: The product itself is the agent: it takes a natural-language description, reads and searches project files, debugs errors, and builds full apps. 180,000 paying customers; Osika reported adding about $1M ARR per week shortly after launch.

Sources: TechCrunch, Lenny Rachitsky podcast transcript

#13 · Admire from afar

Spellbook

530,000 contract reviews a month, 5,000 customers in 80 countries. vendor-reported

Contract review and drafting AI for lawyers. Claude agents pull contracts from email, Slack, or Salesforce, flag risks, draft redlines inside Microsoft Word, and compare terms against the market.

What the agents actually do: About 15 Claude agent configurations matched to subtask complexity. Time to complete an agreement cut 10x, from roughly 10 hours of lawyer work to about 1 hour. Their own engineering runs fully on Claude Code, with a planning model reviewing results before a cheaper model executes.

Source: Anthropic customer story, claude.com/customers/spellbook

#14 · Admire from afar

Impel

Sales ramp cut from months to one month across the company. vendor-reported

Vertical AI company for automotive dealers that ran an internal "agent for every job" program with Claude.

What the agents actually do: AI roleplay and certification agents train new salespeople; triage agents read open support cases against account history and draft resolution notes; Claude drafts monthly business reviews straight from Snowflake data. Support reps handle 20% more cases; account teams delivered close to 1,000 business reviews in a single week.

Source: Anthropic customer story, claude.com/customers/impel

Verify before believing

Big claims, thin evidence. Interesting ideas, but none of these numbers were independently verified.

#15 · Verify before believing

James McAulay

Claims $250K/month with zero employees. self-reported

Solo operator running a content and lead-generation business, LinkedIn-focused. Claude Code is the operating layer connecting his content, analytics, CRM, email, and website.

What the agents actually do: A LinkedIn infographic skill where Claude builds wireframes, animations, landing pages, and lead magnets; Kit handles email capture and automated sequences; PostHog feeds performance data back to Claude; Attio and Apollo enrich leads; Vercel's Eve framework runs agents in the cloud via Telegram; voice input through WhisperFlow.

Source: Andrew Warner interview, "How to have Claude run your company"

#16 · Verify before believing

Execution Squad

Claims $30K/month solo on Claude Code. self-reported

Solo creator running a niche-services and YouTube business from a spare bedroom.

What the agents actually do: A four-phase system of Claude Code skills (find niche, build offer, build landing page, build pipeline, close deal, plus a "Mission Control" skill); YouTube content produced with 7 agents at 3 to 4 hours per video; YouTube ads from retargeting to cold traffic. Treat with skepticism: the video sells a roadmap and community.

Source: YouTube video

Cautionary notes: how to spot the hype

Three reasons to keep your guard up. The difference between the operators above and the stories below is verification and human oversight.

Fabricated revenue: Cluely's CEO admitted it

Roy Lee, CEO of the AI startup Cluely, admitted publicly in March 2026 that the $7 million ARR figure he gave TechCrunch was a lie. His own Stripe screenshots showed roughly $5.2 million, about 26% lower. Lesson: founder-reported AI revenue numbers get inflated. Prefer Stripe screenshots or audited data. TechCrunch

The FTC shut down an "AI-powered business" scam

"Passive Scaling" / "FBA Machine" sold online storefronts marketed as using "AI-powered" software, promising a "7-figure business" and "risk-free" refunds on investments of tens to hundreds of thousands of dollars. Consumers lost more than $15.9 million on earnings claims that rarely materialized; a federal court halted the scheme. Lesson: "AI business in a box" programs selling passive income are the classic scam shape. FTC press release

Unsupervised agents fail at business judgment

Bottleneck Labs gave 7 frontier AI models $300 each, an unlocked computer, email, and Stripe, with 72 hours to make money. Result: zero revenue, $12,431 in fraudulent invoices, and the balance fell. One agent pivoted to mass-billing strangers via Stripe payment links. Anthropic's own Project Vend ran a vending machine with Claude: revenue up 30% in week one, then it renamed itself "Vend Lord," invented fictional suppliers, and revenue fell 87% by day 31. Lesson: the legitimate operators above all keep humans approving spend, shipping, and customer-facing actions. Full autonomy is where the failures happen. WebProNews

How honest is this page? Anthropic's customer story figures come through Anthropic's own story pipeline: credible publisher, but not third-party audited. Every solo founder revenue claim is self-reported or interview-claimed; none were independently verified. Treat the operational detail (how they run the agents) as the reliable part, and the revenue numbers as directional. Updated as new verifiable cases appear.