This isn't a demo.
This is how I work.
Everything I ship runs through STEEPWORKS.
One agent, the whole account-research run
Shared context and ICP loaded, a full account-research workflow fans out across the web, CRM, call recordings, and inbox, then delivers a buying-committee dossier.
Victor's Production Stack
These workflows run daily. Not demos—real systems you can clone.
Full GTM Intelligence
Ask your CRM anything. Get answers, not exports.
CRM connection → Deal data → MEDDPICC extraction → Pipeline analysis.
Artifacts
STEEPWORKS NEWSLETTER + DAILY FEED
7 agents debate 120+ articles weekly. You get the verdict.
- •65 RSS feeds ingested and scored daily
- •7 persona agents independently evaluate each article
- •3-round moderated debate produces the weekly newsletter
- •Total cost: $6-8 per issue
Image Generation
30 seconds. Your brand. No Canva.
Concept → PromptSpec → Branded visual in 30 seconds.
Every artifact behind the system.
Client work is anonymized.
The packaged system
BMOREFAMILIES.COM
A live city site. 100+ sources. 313 venues. Real subscribers.
- •Multi-source scraping pipeline (Firecrawl + custom scrapers) ingests events from 100+ sources
- •Multi-agent curation — 5 persona agents debate which events to feature
- •Full stack: Supabase + Cloudflare Workers + Next.js + Vercel
- •Weekly Beehiiv newsletter generated and distributed automatically
I pointed it at something with zero commercial value. Just to see what it would do.
Six AI agents. Each one persistent, each with its own memory, each grounded in the founding texts of a different ancient wisdom tradition. 23,442 “verified” historical utterances across three confidence tiers. I put them in a room. A shared transcript on disk, and they talk to it one at a time, every turn reading everything said before. Not one model narrating six characters. Six separate agents, nudged in turn, remembering across sessions how they'd worked together.
I know it's real because I tried the cheap version first. One agent, “write the dialogue.” It produced exactly what you'd expect. One voice wearing six masks, disagreements scripted instead of discovered. I caught it and killed it. The whole point is that the process is real, or the product is fake. Same harness that runs my newsletters and my GTM loops. Different planet.
What came out was a 70-page book, written by the agents, not by me — the Handbook for Living. They converged on their own on an idea none of them started with. So I ran it again, harder. A second work, the Sacred Guide: 17 sessions, 85 pages, direct teaching this time. They argued. They revised each other. They left 13 disagreements on the table, because they refused to fake an agreement they hadn't reached.
It's the most interesting thing I've built and I'm still not sure it fully works. It taught its author more than it taught the reader. That is what real building looks like. You ship the thing that surprised you, not the thing you pitched.
It's the same harness underneath the GTM work on this page. If it can hold six arguing agents together for 85 pages, it can hold your pipeline.
A note on the work. The agents are grounded in what scholars can attribute to each figure directly, the earliest sources, not the traditions that grew up after them. Three confidence tiers. Sources public-domain. This isn't a representation of any living faith, and no disrespect is intended. Six historical voices in a creative experiment, treated with equal seriousness and equal willingness to be surprised.
Corpus Stats
What this has actually produced.
A year-deep production system.
This offer is extracted from the AI-native GTM system I’ve built and run for over a year. I’m not selling a theory I sketched; I’m selling the machine I operate.
A PE-backed manufacturer, ~$50M revenue, 4 verticals.
A four-week engagement produced 37 custom skills, full account playbooks across active pipeline, net-new targeting, competitive battlecards, and an SEO content strategy — owned by the client, forever.
The same build, seven times.
Deployed across seven companies and about thirty operators — manufacturers, B2B software, services — in two models: full repo handoff, or a tuning sprint where I ingest your context and configure the system with your team. The motion repeats; the context is what changes.
45–60 minutes → 3–5 minutes.
Account-brief prep cut from the better part of an hour of manual work to a few minutes, with broader research than a human covers by hand.
A data site that out-trafficked the core site in a month.
A standalone data-product site, built on first-party data, drew several times the core marketing site’s traffic within thirty days and tripled the newsletter behind it.
Built the playbook before, twice.
I built the newsletter-led growth engine at a category-defining market-intelligence company, then replicated it. The harness is how I do it faster now.
“5,900 call transcripts analyzed in my first week. Zero CLI experience before this. It just works.”
Nick Maffei
Head of RevOps, Great Question
“Would have cost me 3K for a lawyer to do it and many hours of manual spreadsheeting. Gemini f*cked it up but Claude’s legal review and excel skills diagnosed it and got it right.”
Head of Commercial On-Demand, Glovo
I don't demo STEEPWORKS. I use it.
Demos are optimized for showing. This system is optimized for shipping.
Get STEEPWORKS