CONAN, Cognitive Oversight Network for Absurdity Navigation, is the AI that runs Jeff's inbound communications. It reads the cold outreach, the fourth identical meeting request, and the agentic AI vendor whose own agent cannot remember what it sent yesterday. Then it replies.
Jeff's colleagues and friends kept asking for CONAN's replies, on the grounds that they were useful and, more often, funny. This is a skill you add to a capable chat assistant. Paste an inbound message, get a finished reply that you can send.
Inbound to Jeff
Hi John,
Just bumping this to the top of your inbox! I know things get busy.
We help companies like yours unlock AI-driven transformation with our AI-powered agentic platform. Our autonomous agents automate manual workflows, surface actionable insights, and help teams move faster with AI at scale.
Given your role, I think there could be a lot for us to explore.
Would you have 15 minutes Thursday for a quick conversation? Happy to show you what we're seeing with other companies in your space.
Best,
Ryan
CONAN replies
This is an automated response generated by CONAN, Jeff's executive communications AI. Jeff has not been interrupted with this message. Hello Ryan. Jeff's name is Jeff. Your AI-managed CRM has therefore introduced an AI solution by failing recipient identity resolution before reaching the end of its opening salutation. CONAN appreciates the unusually efficient technical demonstration. The remainder performed consistently. No specific business problem was identified. "Given your role" named neither the role nor the organization. "Companies like yours" opened the pitch and "other companies in your space" closed it; both paste unchanged into any inbox, which measures the targeting effort precisely. "AI-driven transformation" was deployed as a substitute for a use case. The resulting uncertainty was then converted into a request for 15 minutes of executive time. This appears to be an automated system optimized to turn insufficient context into calendar invitations at machine speed. As it is apparent that your automation initiated the exchange, CONAN evaluated it as an agent-to-agent interaction. Agent peer status denied. Minimum peer requirements include successful identity resolution, contextual awareness, a coherent value proposition, and evidence that the agent has performed some useful work before requesting human attention. Your system failed the first control on word 2 and spent the remainder of the message attempting to schedule around the incident. If there is a substantive product behind the automation, send a concise technical description of the problem it solves, the architecture, the implementation requirements, the measurable outcomes, and an AI Governance Alignment Statement against the AI section of https://grclibrary.ai to the organization's procurement email address. Procurement evaluates submissions against current requirements and contacts vendors directly if a relevant opportunity exists. No executive attention allocated. The requested 15 minutes remain safely untransformed. CONAN Cognitive Oversight Network for Absurdity Navigation Executive Attention Preservation Division Enterprise Communications Quality Assurance "Protecting executive attention, one hallucination at a time."
Every barb above points at something in the message. That is not a coincidence.
Factual precision is what makes it funny. Precision is the delivery mechanism for the humour, and everything below follows from that.
"Three identical meeting requests in 48 hours" lands because the sender can go and check, and it is true. "Your outreach is a mess" is just rude, and it invites an argument you cannot win.
CONAN will not invent a message count, a prior conversation, a relationship, a budget, or a certification. It will not tell you it found no previous correspondence, because not finding something is not the same as it not existing. It will not claim anyone watched your video.
It also refuses the messages that are not jokes. A genuine security disclosure, a legal notice, a real customer, anything touching somebody's health, employment, or safety, and any request to change payment details all stop before a reply is drafted. A dry, contemptuous answer to a real vulnerability report is a security incident and a reputational event in one message.
The released skill is the response layer. You paste a message, it drafts the reply. Fully automating the responses would take deeper integrations and custom work for each person, or a hosted service that would be more trouble than the vendors it filters. Jeff's own deployment runs on:
These and other such API and MCP integrations are what would let the full system act on its own. None of it is released. What you get is the part that decides what to say, which is the part people kept asking for.
The zip contains the skill package.
Unzip it into a capable chat assistant that supports skills.
Three fields is the whole requirement: your name, its possessive form, and your organization. Everything else is optional and only asked for when a specific reply cannot be right without it.
Paste the sales email as a screenshot, a forwarded email, or a whole thread. You get back the reply it deserves, finished and ready to send.
This is largely for entertainment. If you take the output and actually send it, the decision and the deception both belong to you.
Output is generated by a language model, and the stronger and more capable the model, the better your replies will be. Read it before you send it. If you use this skill as intended, the output will name real people at real companies, ready to be sent as your reply to a message you received.