According to Andrew Erickson, roughly how much does it cost per month to run an AI agent on a basic VPS, and how much RAM does a single Hermes agent typically need?

Answer at bottom of email

Andrew Erickson's first real business had nothing to do with software. He built a hobby brand selling beer cap maps, decorative wall art made from bottle caps arranged into the shape of a state or country, sold mostly through Etsy to hobbyists and craft beer fans. It sounds small because it was small, at least at the start.

But Andrew treated that hobby brand the way an engineer treats a system to be optimized. A former Qualcomm employee by training, he approached the beer cap map business with the same rigor he might have applied to a chip design problem, hacking together production processes using maker-space equipment most competitors never considered.

That approach eventually turned a niche Etsy hobby into a seven-figure exit in the years that followed, a wildly outsized outcome for a product category most people would dismiss instantly as too small to matter.

He did not stop there. Andrew built a second brand around Montessori-style educational toys, applying the same systems-thinking approach, and sold that business as well. Two exits from two categories that had nothing to do with his engineering background on paper, but everything to do with it underneath the surface.

Andrew's lesson from three very different businesses:

His current focus channels that same instinct into a new frontier: teaching entrepreneurs to build tiny, single-purpose AI applications, what he calls "logic lead magnets," tools that solve one specific problem for a user in under 60 seconds.

A bath-bomb brand building a bedtime story generator. A fishing gear brand building a zip-code-based gear recommendation tool. Small, focused, and genuinely useful, built not by a development team over months, but by a founder over a weekend.

This week pairs Andrew Erickson's practical framework for building tiny AI applications in minutes with William DeVito's argument that most business owners already have everything they need to work confidently with AI agents, they just have not organized it yet. Both conversations point toward the same shift: building genuinely useful AI tools is no longer gated behind large engineering budgets or specialized technical teams.

Andrew's "logic lead magnet" concept reframes what a lead magnet even needs to be. Instead of another downloadable PDF nobody reads, a tiny AI-built app that solves one real problem in under a minute creates genuine engagement and a reason for a prospect to remember your brand.

William's framework goes a layer deeper into infrastructure, arguing that every business needs five core AI-readiness documents before any agent can actually represent the brand accurately.

Watch These Misfits Episodes to Dive Deeper:

The logic lead magnet strategy reframes what a lead magnet needs to be in an era where AI tools make building small, functional software genuinely accessible. Instead of another downloadable guide or checklist, which most prospects download and never open, a logic lead magnet is a tiny, single-purpose AI-powered tool that solves one real, specific problem for a visitor in under a minute.

Why this works: a functional tool creates an active experience rather than a passive download. A prospect who types their zip code into a tool and receives a genuinely useful, personalized recommendation has just had a direct, memorable interaction with your brand, something a static PDF simply cannot replicate. That interaction also naturally captures data (an email address, a zip code, a stated preference) that a generic lead magnet form rarely gets as willingly.

How to build a logic lead magnet in practice:

The payoff is a lead magnet that actually gets used and remembered, rather than downloaded once and

forgotten. In a landscape where every brand offers some version of a free guide or checklist, a genuinely useful tiny tool stands out simply by being different, and by actually working.

As logic lead magnets and other small AI-built tools become more common, a new discovery challenge emerges: making sure AI agents and answer engines can actually find and correctly describe what your tool does, not just human visitors browsing your website directly.

Structure your tool's landing page with extremely clear, plain-language descriptions of exactly what problem it solves and what inputs and outputs it requires. AI agents and answer engines parsing your page need explicit, unambiguous language, not clever marketing copy that requires inference to understand the tool's actual function.

Add structured data markup describing your tool as a specific type of application or service, following schema.org conventions where applicable. This gives AI systems a machine-readable signal about what your tool does, separate from the human-facing marketing copy on the same page.

Consider whether your tool should expose an MCP-compatible interface, allowing AI agents to actually interact with and use the tool directly on behalf of a user, rather than only being described to a user who then has to visit your page manually. This is an emerging discovery layer that early adopters are already building toward.

Document your tool clearly in any developer or API documentation you publish, since AI systems increasingly reference technical documentation directly when determining whether and how to recommend a tool to a user asking a related question.

For help with email marketing for your brand, contact dragon.fish

The Brand: Vessi, the waterproof knit sneaker brand, built its early growth almost entirely on a single, highly demonstrable product claim: shoes made from a knit material that could stay completely dry even when submerged or exposed to heavy rain. Rather than relying on abstract lifestyle branding common in the footwear category, Vessi built its marketing around direct, visual proof of that one specific claim.

The Strategy: The strategy worked because the product's core benefit was inherently demonstrable on video, a shoe getting soaked in a bucket of water and coming out completely dry on the inside is immediately convincing in a way that abstract comfort or style claims are not.

This gave the brand an enormous advantage in short-form video advertising, where a clear, visual proof point outperforms a subjective style claim almost every time.

The Result: Vessi grew from a crowdfunded startup into a brand generating tens of millions in annual revenue within a few years of launch, driven substantially by paid social campaigns built entirely around demonstrating that one core waterproof claim repeatedly across different scenarios and use cases.

The Lesson: If your product has one genuinely demonstrable, visual claim, build your entire early marketing strategy around proving that single claim as clearly and repeatedly as possible before diversifying into broader brand messaging.

The soil a tobacco plant grows in shapes its flavor as much as anything a roller does afterward, and few regions demonstrate this more clearly than the Estelí valley in Nicaragua. The volcanic soil found throughout the region carries a mineral density that many growers and blenders credit for the bold, peppery character found in Nicaraguan tobacco compared to leaf grown in other countries.

Estelí sits in a valley surrounded by mountains that create a distinct microclimate, with enough consistent rainfall and sun exposure to support multiple tobacco varietals within a relatively compact growing region.

Many of the world's most recognized cigar blending families source at least part of their filler and binder tobacco from this exact area, treating the specific plot of land almost the way a winemaker treats a single vineyard.

A cigar blender working with Estelí tobacco is not just choosing a flavor profile, they are choosing a specific patch of volcanic earth that cannot be replicated anywhere else on the planet, which is exactly why so many premium blends proudly print the growing region right on the box.

The next Collective Minds Society Cigar & Whisky trip is Feb 18-22, 2027

Use an AI coding agent to build a simple, custom dashboard that pulls together scattered e-commerce metrics into one view, without needing a dedicated developer.

Sample Prompt:

"Build a simple dashboard that pulls daily ad spend, email revenue, and inventory levels from these three platforms into one view, with an alert if inventory drops below 100 units."

A custom dashboard built in an afternoon beats checking five different login screens every single morning.

Q: What is a "logic lead magnet"?

Andrew Erickson describes it as a tiny, single-purpose AI-built app that solves one specific problem for a user in under 60 seconds, replacing the traditional downloadable PDF lead magnet.

Q: What are the five core documents William DeVito says every business needs before deploying AI agents?

A business knowledge base, a brand expert voice document, a style guide, customer personas, and marketing pillars.

Q: How much does it cost to run a basic AI agent infrastructure?

Andrew Erickson notes a basic virtual private server runs roughly ten dollars a month, with each agent typically needing only two to four gigabytes of RAM to operate effectively.

Q: When should I use Claude's Sonnet model versus Opus?

William DeVito suggests Sonnet handles tasks completing in under an hour well, while Opus is better suited for longer, more complex tasks that take over an hour to complete.

A basic VPS to run an AI agent costs roughly ten dollars a month, and a single Hermes agent typically needs two to four gigabytes of RAM.

Andrew Erickson made the point that the infrastructure cost of running genuinely useful AI tools has become negligible for even a solo founder.

Watch the full conversation