
Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
The business is selling an “AI employee,” not AI tooling — Nick says the winning offer is a frictionless $5,000/month service with unlimited agents, usage, monitoring, support, and ongoing changes, because executives don’t want to hear about tokens, credits, or infrastructure.
You can start broad, but the money is in verticalizing around executive pain — He recommends starting with markets like marketing agencies, law firms, insurance agencies, manufacturers, wholesalers, and real estate agencies, then narrowing into sub-niches once you see pull, since most buyers share the same headaches: too many emails, meetings, follow-ups, and open loops.
The stack is opinionated: Granola, Trello, Loom, Claude Code/Codeex, Hermes, Orgo, Composio, AgentMail, and Obsidian — Nick’s core claim is that Obsidian acts like a real “second brain,” Composio solves the authentication/security nightmare, and Hermes beats OpenClaw on reliability because it’s more self-evolving and model-flexible.
Agents should build agents — Instead of manually configuring every customer environment, Nick uses one Hermes agent over Telegram to spin up and manage other Hermes agents inside Orgo VMs, calling this the unlock that lets a solo operator support dozens of customer setups.
Reliability is the real product after the sale — His fulfillment advice is to get the first agent live within 48 hours, limit customer requests through a Trello board to avoid scope creep, and add watchdogs plus alerting so you know when a gateway, cron job, or skill breaks before the client does.
Content is the unfair distribution advantage for this kind of solo business — Greg literally found Nick through an Instagram clip about OpenClaw, and both argue that publishing consistently is the highest-leverage way to warm up leads, earn trust, land podcasts, and avoid selling cold.
Nick opens with the core pitch: businesses will pay $5,000 a month for a managed AI agent setup if you remove all the friction. His framing is sharp — don’t sell “agents,” sell a digital employee that knows the business and improves every week. The trick is offering “unlimited” agents and usage even though, in practice, most clients only need one to three agents, not 10 or 100.
He steers people away from heavily regulated starting points like healthcare and finance, and toward marketing agencies, law firms, insurance agencies, manufacturers, wholesalers, and real estate agencies. Greg adds that these are all people-heavy businesses, which means lots of inefficiency and lots of room for automation. Nick’s nuance: you don’t have to start hyper-niche on day one — test a few markets, then narrow once one starts pulling you in.
Across industries, Nick says the actual buyer is usually an executive or decision-maker drowning in emails, meetings, follow-ups, and context switching. That becomes the wedge: start with a reusable executive-assistant style agent, then layer in vertical skills like case management and demand letters for a law firm. His point is that the top of the org chart has similar pain almost everywhere, even if the domain workflows differ.
Before getting technical, both of them pause on distribution. Nick says the best sales call is one where the prospect already knows who you are and what you sell, and Greg backs it up with the story of discovering Nick via an Instagram video while doom-scrolling at midnight. Their takeaway is blunt: content is “overpowered,” and in an AI world, it’s the highest-leverage way to warm up leads, build trust, and make everything else easier.
Nick’s customer-facing workflow starts with Granola for meeting notes, which feed into Trello as client requests on a kanban board. That board becomes the scope-control mechanism: clients can ask for improvements, but he tries to constrain it to one or two requests every 48 hours so fulfillment doesn’t spiral. Loom handles constant progress updates, while Superhuman and Asana help him stay fast on email and internal operations.
For actually building agents, he likes Claude Code and OpenAI Codex as builder tools, but recommends Hermes as the thing clients actually use because it’s more reliable and lets you swap models as better ones arrive. Composio is his must-have connector layer for thousands of apps and secure authentication, AgentMail gives each agent its own email identity, and Obsidian is the star of the show — a markdown-based “second brain” he’s built since November 2025 that gives agents long-lived memory across people, projects, and daily transcripts.
The live demo centers on Orgo, where each customer gets a workspace and each agent gets its own cloud computer. Nick’s argument against local Mac minis is practical: cloud VMs are easier to access, isolate, replace, and secure, especially when you’re managing lots of clients. The bigger unlock is recursive — his Telegram-based Hermes agent uses the Orgo MCP to spin up, configure, and maintain other agents, which is how a solo operator can scale.
Nick closes on reliability, which he treats as the real business. Gateways can crash, cron jobs can fail, and tools can break, so he sets up watchdogs for auto-restarts and has agents email him when something goes wrong. That’s the difference between a cool demo and a durable service business: clients become dependent on the agent, so your job is to make sure it quietly keeps working.
Share
Keep Reading
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.

Playbook
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.

Playbook
OpenAI shipped /goal for the Codex CLI. It turns a prompt into a persisted, self-continuing contract.