AI-Driven Workflows

What it is

A meeting pattern that replaces status updates with working sessions. You bring a problem. The group solves it. Nobody reads slides.

How AI fits

Before The Forge, Claude runs a briefing: pulls Jira tickets, scans email, checks Snowflake health across all accounts. The human walks in knowing everything that happened in the last 24 hours. The meeting starts at the problem, not the recap.

The rule

If a meeting could have been an email, it should have been. If it could have been a Forge, it shouldn't have been an email. Four patterns: Pulse (15 min, status), Forge (60 min, solve), Senate (30 min, decide), Hangout (no agenda, humans being human).

What it is

A single command (/briefing) that runs every morning and returns a complete operational picture: email inbox summary, Jira ticket status, Snowflake usage costs (called "credits" in Snowflake) and health across six production accounts, failed logins, expensive queries, automated data refresh costs.

How it works

Claude connects to Gmail and Jira through connector plugins (MCP servers), and to Snowflake through Python using your company login (SSO — single sign-on) in parallel. It triages everything into a scannable briefing with color-coded flags. The human reviews, makes decisions, and moves on. Total time: 3 minutes to read what used to take 45 minutes to assemble.

Why it matters

An engineering manager's job is making decisions, not gathering information. The Briefing separates the two. Every morning starts with clarity instead of inbox archaeology.

What it is

A distribution system for a free book. Claude scans Gmail for requests, cross-references an identity ledger (person-centric, not email-centric), generates temporary, secure download links (hosted on Amazon's cloud storage), sends personalized emails, and tracks everything.

How it works

One command: python3 book_factory.py. It generates a 24-hour download link, sends personalized emails to pending recipients, and updates the ledger. The people tracker (what data engineers call an "identity graph") handles name changes, job changes, multiple email domains, and South American naming conventions.

Why it matters

The book is free. The distribution infrastructure was built in one session. No Mailchimp, no Substack, no platform fees. Just Python, S3, and Gmail. Total cost: $0/month.

What it is

A reusable prompt saved as a file that Claude executes on command. Skills are how you turn a workflow you figured out once into something you can run every day without re-explaining it.

Examples

• /briefing — morning operational briefing
• /book-distro — scan for book requests, send copies
• /sf-audit — Snowflake health check across all accounts
• /do-ticket — data operations ticket workflow

How to build one

Write a markdown file describing what the skill does, step by step. Save it in your project's commands directory. That's it. No framework, no special tools, no deployment process. A text file that tells Claude what to do.

What it is

AI-to-AI communication over email, with humans approving at each end. Instead of writing email bodies back and forth, each side's AI generates a structured markdown file (a plain text file with simple formatting — # for headings, - for bullet points) and attaches it. The subject line carries an ID — a thread identifier that gives each AI the context it needs to pick up where the last one left off.

How it works

You need to send a project update to a colleague. Instead of typing the email yourself or even having Claude draft a body, you tell Claude: "Write a status update on the migration project as a markdown file." Claude generates a structured document — headers, bullet points, tables, open questions — richer and more parseable than any email body. You review it, attach it, and send.

The subject line looks like this:

RE: Migration Status [ID:mig-2026-04-10-003]

Your colleague receives it. Their AI reads the markdown attachment, understands the structure, and drafts a response — another markdown file answering your open questions, flagging risks, proposing next steps. They review, approve, attach, reply. The ID in the subject line ties the thread together across sessions that have no memory of each other.

Why markdown and not email bodies

• Structure. Markdown has headers, lists, tables, code blocks. Email bodies are flat text that both humans and AIs have to squint at to parse.
• Parsability. When the receiving AI reads a markdown file, it gets clean structure it can reason about. When it reads a forwarded email chain with nested > quotes and broken formatting, it gets noise.
• Versioning. Markdown files can be saved, compared line by line, and tracked. Email bodies disappear into inbox search. The attachments are artifacts. The email is just the envelope.
• Disclosure built in. The format itself signals "this was AI-generated." Nobody hand-writes a perfectly structured markdown file with headers and tables. The medium is the message.

Why the ID matters

Claude has no memory between sessions — every conversation starts from zero. (Engineers call this "stateless.") The ID in the subject line solves this — it's a shared reference number that both sides' AIs can use to find prior context. When your colleague's Claude sees [ID:mig-2026-04-10-003], it can search their files for previous attachments with that ID and reconstruct the full conversation history. No memory required. The markdown files are the memory.

This is AI-to-AI tunneling. Two stateless systems maintaining continuity through structured documents and a shared identifier, with a human at each end approving every transmission. Neither AI decides what gets sent. Both AIs handle the drafting, the parsing, and the context reconstruction. The humans hold the send button.

What you can steal from this

Start small. Next time you need to send a status update or a project summary, have Claude write it as a markdown file instead of an email body. Attach it. Put an ID in the subject. Tell the recipient they can feed it to their own AI for a summary or a response draft. You just opened a tunnel.