Agenteers in the Wild
Fourteen verified Agenteers and the tools they personally built, across finance, law, music, healthcare, marketing and tech. The fact-checked companion to 'What is an Agenteer?'.
This is the companion piece to "What is an Agenteer?". The essay defines the role. This article shows the people.
Every profile here is someone non-technical who personally built the thing — not someone who sponsored it, led a team that built it, or wrote the spec and handed it to engineers. We checked. Where a primary source showed the person led rather than built, we removed them, even when the story was good. The list is shorter than it could be on purpose. Fourteen people we can stand behind beats fifty we can't.
How we verify: every profile is anchored to a primary source where the person describes building it in their own words, or an independent account that names them as the builder with no engineering team doing the actual construction. Vendor case studies are corroborated against a second source where possible. If you're building and want in, send your story.
Professional services & law
Jamie Tso — Founder, LegalQuants (ex-Senior Associate, Clifford Chance)
Jamie was a Senior Associate at Clifford Chance, one of the world's largest law firms (Magic Circle), until he resigned in April 2026 to found LegalQuants. He's not an engineer. He experimented with TensorFlow at one point and is candid that it "didn't go very far" technically. His daily life was law, not code.
The why: Big Law runs on repetitive, high-stakes document processing — redlining contracts, building signature packets, mapping fund prospectuses for due diligence. The mechanics are pattern-matching. Jamie started building tools for himself, then for colleagues who kept asking how he was shipping work faster.
What he built:
- RedlineNow — instant contract redlining, built in 2 days (December 2025).
- SignaturePacketIDE — automated M&A signature packet generation, used in real BigLaw deal teams.
- A fund prospectus mapping tool — his document tools went viral internally and were adopted by the firm.
The stack: Microsoft Copilot Studio and Power Automate internally; "vibe coding" with open-source AI tools externally. His public GitHub (jamietso) shows the commits.
The angle: He asks the AI to build a deterministic tool rather than answer questions live inside the workflow, which lowers error rates. The tool is reliable; the AI is the variable he controls during the build.
"I got obsessed and started building not just for myself but for colleagues."
Read the Artificial Lawyer interview · GitHub
Traditional industries
Tyler Diogo — Operations Manager, Arden Insurance Services
Tyler is Operations Manager at Arden Insurance Services, a traditional insurance MGA. His background is operations and accounting. He found automation by accident, running a Discord community for a video game on weekends — he needed Zapier for Discord notifications and calendar updates.
The why: Arden was sending tens of thousands of past-due invoice notifications a year by hand. Reps chased customers for payment; legal mailings were a manual queue. Tyler realized the same Zapier patterns from his hobby could rebuild the whole AR workflow.
What he built: A 12-Zap workflow chain for past-due notifications and legal mailing. It worked, and automation became the company's operating philosophy. Tyler now runs Arden's broader automation program — billing, customer communications, internal reporting, compliance.
The stack: Zapier, with AI logic embedded in some workflows.
The numbers: 17,000+ hours of work automated in 2023, projected to roughly double in 2024 (the often-cited "34,000 hours" is the 2024 projection, not a settled figure). $500,000+ in annual overhead savings. ~$150M in annual company billings flow through automated processes.
"Our customers now chase after us to pay their invoices, rather than us chasing after them."
Read the Arden Insurance story
David Gabriel — VP Marketing, Rhumbix
Rhumbix is a 12-year-old construction technology company recently acquired by Autodesk. David is VP of Marketing, a career marketer, not an engineer. Rhumbix ran a deliberate AI-native transformation under CEO Zach Scheel. David's sales build is one piece of that.
(Primary source for this profile: Kasper's recorded podcast conversation with David, plus the Lindy case study. Specific agent-count and tooling detail come from that first-hand interview.)
The why: Sales reps were juggling three tools (n8n, Make, Zapier) plus mobile workflows for prospect research, meeting prep, and follow-up. Cost was rising; the experience was fragmented. David wanted one system end-to-end.
What he built: A Lindy AI Assistant deployed across the Rhumbix sales team — email triage, meeting prep, proactive scheduling, mobile workflows via text, deal tracking, meeting-notes review. Roughly five-minute setup per user. No engineering involvement on the build.
The stack: Lindy as the orchestration layer, connected natively to Salesforce, Gmail, calendar, and the sales stack.
The numbers: ~$25,000/month in cost savings vs. manual workflows. A large share of go-to-market work delegated to AI agents. Three previous tools consolidated. Full sales team onboarded within days.
"Deploying Lindy felt like flipping a switch. I don't know any AI agent that has taken me five minutes to build, even using Claude Code."
Caren Kelleher — Founder, Gold Rush Vinyl
Caren founded Gold Rush Vinyl, a vinyl record pressing plant in Austin, Texas, running 24-hour production with a small team. Her background is partnerships at Google (Head of Music App Partnerships) and a Harvard MBA. Zero coding background — she found Zapier by Googling how to move a PDF.
The why: A pressing plant has a chaotic operational layer: order forms, file delivery, production scheduling, status updates, shipping. Without automation Gold Rush couldn't grow, and Caren needed a tiny team to run like a much bigger one.
What she built: Over time, 76 active Zaps automating 18,000+ tasks per year — order intake, customer onboarding, file QA, production handoff, status notifications, shipping. Zapier's own write-up states plainly: "Caren built each step of these critical processes within Zapier."
The stack: Zapier, integrated across Google Workspace, Stripe, support tools, and the production system.
The numbers: 2,285 hours of manual work saved per year (≈1.2 years of labor), letting a deliberately small team stay focused on records and musicians.
"We couldn't do our work without Zapier. The work it's doing helps us do more work for musicians."
Read the Gold Rush Vinyl story
Tech & startups
Ondrej Machart — Head of Product, Livesport (Flashscore)
Livesport is the Czech media company behind Flashscore, a sports data platform with 100M+ monthly users. Ondrej runs core product. His background is product and UX design, not engineering. He started using Claude Code in September 2025 and documents the journey publicly.
The why: Like every product head, Ondrej had a backlog of internal tools that would help his team but never made the roadmap. Engineering was reserved for the public product. With Claude Code he stopped waiting.
What he built (from 13 projects in six months): a Product Portfolio Coach that informed a C-level brand consolidation; a reusable iOS prototyping baseline; an editorial infographics tool; a data-presentation system used for major decisions; and vibe-coding workshops that taught colleagues to ship their own internal tools.
The stack: Claude Code as primary. He hands off to senior engineers when something needs production QA — and says so openly.
The framing: "This isn't magic, it's work. AI gets you far, but it isn't free, effortless, or risk-free."
Aditya Vempaty — VP Marketing, MoEngage
MoEngage is a Series F customer-engagement platform. Aditya leads marketing for North America. Important nuance: he has an engineering education (Georgia Tech) and early-career time at Nutanix and Dell/EMC, but he has spent his career as a marketer, not a software developer. He is not the "never touched tech" archetype — he's a marketer who builds.
The why: AI-generated marketing copy sounds like AI. Aditya wanted to systematically catch and rewrite those patterns.
What he built: A /humanizer skill in Claude Code that scores drafts for "AI-ness" across four categories, flags the patterns, and rewrites them in his voice. He built and iterates it himself.
The stack: Claude Code (skills feature).
The framing: His value is a candid take that cuts against the hype — he's tested most AI marketing tools and concluded some do real work and others are vapor.
"Claude Code does a job. Chat does a job. I honestly don't understand what Cowork is for."
Hiba Fathima — Marketing Specialist, Firecrawl
Hiba runs full-stack marketing at Firecrawl, an AI tooling startup. Honest caveat: she has a CS degree, which she says made the Claude Code learning curve easier. She sits at the technical end of the Agenteer spectrum. We keep her in because her work was previously gated entirely by designer and developer time, and she removed that queue herself.
The why: Small marketing teams hit a wall — every page or campaign needs designer time, developer time, a dependency queue.
What she built: an autonomous brand-mention tracker (runs every Monday, scrapes Reddit/HN/Twitter/forums); and direct landing-page shipping that bypasses the designer/developer queue — she drafts, Claude Code builds, she ships.
The stack: Claude Code as the daily driver; Firecrawl's own scraping API for the tracker.
"If I know exactly what needs to go on a page, I can just... ship it. No waiting on anyone."
Adam Schoenfeld — CMO, Inflection
Adam was CEO and co-founder of Keyplay until Inflection acquired it; he's now CMO at Inflection. A serial SaaS operator, he skipped the standard 30/60/90 onboarding and built an AI-native CMO operating system from day one — the actual way he runs his job.
The why: Most new execs spend months assembling working knowledge in their heads. Adam decided that work should live in machine-readable form from the start, where any agent or future hire could draw from it.
What he built: a Claude Code personal operating system (a three-bucket local file architecture: context, work, system); a /morning skill pulling a briefing from Slack, Notion, Calendar, customer calls, Salesforce, LinkedIn; a structured GTM context knowledge base; and /draft-content + /iterate-content skills with curated voice files. He builds and runs it himself.
The stack: Claude Code as the operating layer; Slack, Notion, Salesforce, Calendar, LinkedIn as sources.
"AI agents are only as good as the context they have. The boring work of organizing and maintaining context is one of the highest-leverage things I've invested in."
Sebastian Volkis — Indie founder
Sebastian studied physics and astrophysics, became a freelance media buyer, then started using AI coding tools out of frustration with the gap between "I have an idea" and "engineers built it." His twin brother Matthew handles marketing only — Sebastian is the sole builder.
The why: His media-buyer brain saw demand patterns engineers wouldn't prioritize. Creators needed a way to find trending viral articles; SMBs needed AI support without enterprise procurement.
What he built:
- TrendFeed — aggregates viral news, generates short-form content. Built end-to-end in 4 days with Cursor.
- ChatIQ.ai — AI support chatbot. Originally Bubble (2023), later Cursor. ~11,000 users, £7.2K MRR (verified via Indie Hackers).
- CodeSpring — his AI coding course, £45K revenue since January 2025 (self-reported, consistent across his public posts).
The stack: Cursor as primary builder; GPT-4 and Claude inside the apps; Bubble for the original ChatIQ.
The numbers: £9,222 (~$12K) in TrendFeed's first four weeks, £5.5K on launch day. He posted a public bank-account screenshot to prove launch-day revenue.
"I built this entire app using AI in just 4 days."
Like this? Get the next one in your inbox before it hits the blog.
Sabrine Matos — Founder, Plinq (Brazil)
Sabrine spent her career in growth marketing in Brazil, embedded in product teams but always on the sideline of engineering. She's not technical. Plinq is her first company. (Independently corroborated via Cybernews coverage and the Brazilian company registry, in addition to the Lovable case study.)
The why: A woman in Sabrine's circle was murdered by a partner with a hidden history of violent convictions. In Brazil there was no consumer-accessible way to check criminal records before dating someone. The gap was lethal.
What she built: Plinq — a women's safety app with instant background checks, green/yellow/red risk scoring, a panic button, and a safety feed. Built end-to-end on Lovable in 45 days — frontend, backend workflows, public-records integrations. No external development.
The stack: Lovable for the entire stack.
The numbers: 10,000+ users within 3 months. R$2.2M (~$456,000) ARR. 300% MoM growth. 200+ dangerous situations identified and potentially prevented.
"If Lovable didn't exist, Plinq would never have seen the light of day."
Nomiki Petrolla — Founder, Theanna
Nomiki spent 15 years in product leadership. She's a non-technical product person who lived the engineering bottleneck for over a decade. Theanna is her solo company.
The why: Women founders face a network and access gap. Nomiki wanted one platform — guidance, community, revenue tracking, a build interface — and was tired of waiting weeks for engineers to ship features she could fully spec.
What she built: Theanna — an all-in-one platform for women tech founders. She personally built the entire user-facing app (every screen, every flow) in Claude Code via natural-language prompts and screenshots, handing off to hired engineers only for backend connections (database, APIs, performance) — a split she discloses openly. The marketing site (12 integrations) was built in Lovable in a single day.
The stack: Claude Code for the production app frontend; Lovable for the marketing site; GitHub; Linear.
The numbers: ~$207K ARR (March 2026), publicly trackable via her building log. 300+ women founders on the platform. Feature shipping cut from ~2 weeks to under 2 days.
"Every single screen, every flow, every interface that the founders on Theanna interact with was built by me."
Healthcare & nonprofit
Arun Nadarasa — PCN Clinical Pharmacist, NHS (UK)
Arun is a working clinical pharmacist in the UK's NHS. His training is clinical, not technical (pharmacy degree; NHS pharmacist since 2012). He spent years stuck at NHS hackathons because he couldn't build — he'd brainstorm with developers but couldn't make anything himself.
The why: NHS clinicians have ideas for tools — triage navigators, blood-test assistants, practice-performance trackers — but no engineers and no way to build. The gap between clinical insight and software was near-absolute.
What he built (since discovering Lovable, September 2024): 20+ healthcare apps without coding — a QOF Progress Tracker, a Smart Blood Test Assistant, a Patient Triage Navigator, and more. He also organized Clinical AI Hackathons drawing 300+ registrants across 7 UK locations (80 sign-ups in the first 6 hours of one event).
The stack: Lovable for everything. "I don't dive into the code. I prompt Lovable with instructions like 'connect to this external API.'"
The framing: He went from hackathon attendee to hackathon organizer in under a year — and now teaches other clinicians to build.
Colin Budries — Head of Support, Truemed
Colin runs customer support at Truemed, a specialized HSA/FSA payment processor. His background is military (US Army Lieutenant) and accounting (Baylor). Not technical in the engineering sense; analytical and structured by training.
The why: Support tickets cluster around HSA/FSA eligibility, compliance, and standard account issues — mostly well-defined answers. The team didn't need more reps; they needed an AI layer to resolve the predictable cases and route the rest.
What he set up: Customer-support AI agents that classify, route, and auto-resolve health-related tickets, compliance-aware, with human escalation for ambiguous cases. He configured and deployed this himself on a no-code agent platform — no engineering support — rather than writing code.
The stack: Lindy as the agent framework, connected to Truemed's support system, knowledge base, and customer records.
The numbers: 5,000+ tickets automated. 36% of total support volume. 67% cost reduction per ticket ($1.00 → $0.33).
"Lindy is like a developer that works just for me."
Justin Steele — Founder, Kindora
Justin spent ten years as Director of Google.org Americas, overseeing nearly $700M in philanthropic giving. He has a Harvard MBA/MPA and a chemical engineering undergrad from UVA, but his career was philanthropy and consulting and he had never built web apps.
The why: Existing grant-matching tools returned thousands of poor matches. Justin had spent ten years on the funder side and knew exactly which signals separated good matches from noise. The expertise was his moat; he needed the artifact.
What he built: Kindora — an AI fundraising platform for nonprofits: funder match scoring, proposal-writing help, funder research briefs, voice pitch practice. The engine filters ~3,000 raw matches to ~75 viable prospects per nonprofit. A hired developer built the early prototype; Justin then adopted Claude Code and has personally built 90%+ of the codebase since December 2025 (he discloses this split openly).
The stack: Claude Code for the build; Claude Sonnet 4.6, Haiku 4.5, Opus 4.6 across assistant, classification, and generation.
The numbers: $100,000 raised in year one for his own nonprofit (mostly cold outreach using his prototype). $101,000 in a single donor month (March 2026). 328 nonprofits onboarded within months of beta. Accepted into 3 accelerators; $50K SAFE from Camelback Ventures.
"It's like jumping into an Iron Man suit. There's such a unique opportunity right now for people who understand their problem space."
Read Anthropic's customer story · Listen to the interview
Market proof — enterprise signals
These are not individual Agenteer profiles. They're organization-level evidence that the pattern scales. We keep them separate on purpose: in each case there's no single non-technical person who personally built a specific tool — these are programs, platforms, or cultures. The real Agenteers inside them are often unnamed (and worth hunting for).
- BBVA (Elena Alfaro, Head of Global AI Adoption). 20,000+ employee-built custom GPTs on ChatGPT Enterprise. A frontline-built Peru ops assistant (used by 3,000+ staff) cut query handling time ~87%. Alfaro builds the conditions, not the tools — the unnamed Peru frontline employee is the actual Agenteer here. OpenAI case study
- Salesforce (Zach Stauber, Sr Director, Agentforce). Profiled by HBR as an "agent manager" — he configures and supervises a fleet of support agents he didn't build. A distinct and real archetype, but a manager of agents, not a builder.
- Walmart — nano agents. A layered "super agent / nano agent" strategy across merchandising, supply chain, service, ops. Org-level; no individual builder named. CEO Doug McMillon: "AI is literally going to change every job."
- Rakuten — Claude Code across departments. Deployed to product, sales, marketing, finance, explicitly beyond engineering. 79% reduction in time to market; bi-weekly replacing quarterly cycles. The non-technical builders are real but unnamed in the source.
- Stanford GSB — AI@GSB. The largest student org on campus (597 members), co-founded by four MBA students including Jenni Steiger (ex-BlackRock chief-of-staff team; the closest individual Agenteer in this entry). Students wire personal/team operating systems from Claude, Cursor, Lovable, and the everyday SaaS layer; internship interviewers now ask MBAs for "portfolios of builds." (Note: earlier drafts cited "~36 AI-integrated MBA courses" and a student trucking-agent → HappyRobot link; both are unsourced and have been removed. Co-founder Celeste Bean is a highly technical ML/hardware engineer, not a non-technical builder.) Poets & Quants
What's the pattern?
Across the fourteen verified profiles, three things repeat:
1. The builder isn't a software engineer. A lawyer, an operations manager, a marketer, a clinical pharmacist, a founder, a media buyer, a support lead. Most had never built production software before. (Where someone has a technical education — Aditya, Hiba — we say so plainly rather than overstate the "non-technical" angle.)
2. The first build is small. A late-invoice workflow. A brand-mention tracker. A landing page. A criminal-record check. A support classifier. The wins compound; the first build is rarely the most ambitious.
3. The tool changes; the skill doesn't. Zapier, Lindy, Lovable, Cursor, Claude Code, Copilot Studio, Bubble. Different stacks for different jobs. The constant is briefing — explaining clearly what you want, iterating until it works, and owning the running system. Spotting, building, and managing it. Not handing it to someone else.
That last line is the membership test. The people who spot it, build it themselves, and run it — that's who this is for.
If this is you, you're an Agenteer. We're building the community at agenteers.ai.
Removed in the May 2026 verification pass (kept for the record): Anri Coetzee (EY), Marie Sanders (EY), Markus Kronen (BMW) — in each case primary sources show the person led or sponsored the work while a named engineering team did the actual building. Strong stories, but not Agenteer stories by our own definition. Rigor over volume.