Agent Native OS
Workshop.
Build agent fleets
that make you money.
You walk out building your Morning Brief, block by block, on a fleet that works while you sleep. Then you keep building it all week and into Summer Camp.
Who's in the room
- ·First time in Claude Code, or early experience.
- ·Here to see what's possible and leave with working blocks.
- ·You build the MVP of every block today.
- ·You live in the terminal. Claude Code is a daily tool.
- ·You want to go deeper, not just get it running.
- ·You customize each block and preview the summer-camp moves.
You did not buy the morning brief.
You bought the architecture.
The morning brief is one expression of what this architecture can do. The repo has hundreds more, and your Claude knows how to install all of them.
Sub-agents, background agents, team agents. All yours.
Blueprints, guides, and a Cairns knowledge vault.
Start it today. Expand it all week and into Summer Camp.
This isn't theory. It's a trade school.
Last cohort walked out with these.
Real morning briefs. Real deliveries. Real machines. By tonight you start yours.
Not professors. Builders.
Co-founder, strategy. The vision + the gains.
Co-founder, lead instructor. The architecture + the build.
PM, teacher, strategist. Keeps the day on the rails.
The LMS + your run-of-show. Where every prompt lives.
GitHub + data. Your go-to when version control bites.
This is the whole system. You bought all of it.
Seven layers, one path: Repo → Harnesses → Orchestration → Agent Native OS → Blueprint → Growth OS → Outcome. Today you install the first layers as your morning brief. Summer Camp takes it all the way to a business that pays.
Everything today lives in one place.
Wade built your portal: a visual map and teacher for the agent-native-os repo. Every prompt, every skill, the full run of show. You have exclusive access. Open it now and keep it open all day.
Here's the day
Today's schedule
| Time (ET) | Section | Node |
|---|---|---|
| 10:30-11:00 | Welcome + Vision | The map: what you're building + why |
| 11:00-11:35 | Sub-Agents + Fleets | First sub-agent swarm |
| 11:35-11:45 | Short Break | 10 minutes |
| 11:45-12:20 | Security + Observability | Keys vaulted + runs traced |
| 12:20-12:55 | Connecting Your Data | Sources wired to the brief |
| 12:55-1:30 | Knowledge + Memory | Cairns 101: teach + demo the layers |
| 1:30-1:40 | Short Break | 10 minutes |
| 1:40-2:15 | Build the One Cairn | Your second brain, on disk |
| 2:15-2:35 | Lunch | 20 min (working lunch + Q&A) |
| 2:35-3:30 | Wire + Deploy: Morning Brief | The deployed brief |
| 3:30-4:00 | Close + Open Q&A | Showcase + what's next |
Doors 10:00 AM ET · wrap ~4:00 PM ET · two short breaks + a working lunch. (PT folks: 7 AM start.)
We build the morning brief
block by block.
Every section follows the same rhythm. Teach the idea, Demo it live, then you Build the block yourself.
The idea, and why it matters. No terminal yet.
We run the skill live. You watch a block get built.
You run the same skill. 4D gets the MVP, 8D goes deeper.
One morning brief. A whole fleet behind it.
Set up Claude Code
like a pro.
No git hell. Everyone starts level.
Get your machine ready
cd ~/GitHub/agent-native-os bash verify.sh # every line green = ready
- Claude Max active
- Claude desktop + Code CLI working
- GitHub CLI authenticated (gh auth status)
- agent-native-os repo cloned
- /personalize run, CLAUDE.md is yours
- (bonus) first connector wired
Leave permission prompts on. Watch what Claude wants to do before you let it.
Auto mode + bypass permissions, with hooks as the safety net.
Build agent fleets
that make you money.
- ·Parallel workers, one context each.
- ·Each gets a scoped job + an output contract.
- ·They all report back to the lead.
- ·Long-running, keep working after you leave.
- ·Good for builds, scrapes, and watchers.
- ·You check the result, not the clock.
- ·A fleet with named roles.
- ·They coordinate toward one goal.
- ·A lead splits the work and merges it back.
Sub-agents · Background agents · Agent fleets
Run a research swarm on your own idea.
Paste the Research Swarm prompt from your run-of-show. It spawns three sub-agents in parallel: web, your repo, your transcripts. Then it synthesizes one consensus brief.
Customize the roles and model tiers (Haiku workers, a stronger synthesizer). Then wire the swarm onto a real recurring job so it runs without you.
The exact prompt lives on the LMS Workshop Day page. No new command to memorize.
Short Break
10 Minutes
You just ran your first fleet.
Next: lock the keys so it is safe to let it work on its own.
How to trust
your fleet.
How To Trust
The Fleet.
Security + Observability. Here is how we keep it safe to let agents work.
It asks before it acts. You promote a project to auto only once you trust it.
Secrets live in Infisical, pulled at runtime. Never in plaintext files.
Version control plus one backup, so any mistake is cheap to undo.
See every agent action, token, and cost. Nothing happens in the dark.
Security and Observability
Lock down your fleet.
infisical login # add your keys in the dashboard infisical run -- claude
- 1.Sign in to Infisical and add your API keys in the dashboard.
- 2.Launch Claude through infisical run so keys load at runtime.
We wire Langfuse live: Stop + PostToolUse hooks that trace every agent action, token, and cost. Watch it in the room, take the pattern home.
- ·Nothing your fleet does happens in the dark.
Your AI can see
your world.
Tools come in two shapes.
A persistent connection: Gmail, Slack, Drive. The fleet reads and writes live.
A one-shot tool the agent calls: fetch, transform, post. Cheap and composable.
Connecting your sh*t
Wire your first source.
- 1.Pick one source you actually use (Gmail or Calendar).
- 2.Wire it in.
- 3.Ask it one real question.
- 1.Wire three sources at once.
- 2.Give each its own sub-agent.
- 3.Have them all feed one brief.
Your fleet
remembers.
Cairns 101: the foundational concept of the whole system.
Every session, your agent starts from zero.
It is not a missing-data problem. It is an untiered-data problem. You either tell it too little or you drown it in everything.
No memory. It forgets yesterday, re-asks what you already answered, repeats old mistakes.
Dump every transcript into context. It drowns in noise, misses the signal, and you pay for every token, every call.
Tiered recall. Load a tiny map always, fetch detail only when the question needs it.
A cairn marks the path.
Tyler stacks stones on hikes with his family. A cairn is that stack: a marker that tells the next traveler which way the path goes.
Not the whole mountain. Just the marker you need to keep moving. One source, one cairn.
Three tiers. Load light, drill deep.
L1 routes. L2 summarizes densely. L3 proves.
~16 categories, ~6k tokens. The map. 60-70% of questions stop here.
Dense 85-135 word summaries. The catalog the map points into.
The raw source, never edited. The exact quote, with provenance.
Every agent asks the Librarian.
The Librarian fans out across every cairn at once. Most questions are answered right here.
Need more? Pull the L2 card. Need the exact line? Fetch L3. It stops as soon as it has enough.
Every answer comes back with its source: which cairn, which tier, which line.
Same answer. A fraction of the cost. None of the noise.
There's an input of information, the repo chunks it, helps build the thing, and then the next agent knows where and what it is.
Hunter
A source comes in. The system pulls it and summarizes it.
It proposes what to remember. Nothing hits L1 without your yes.
Saved, searchable, and ready the next time you ask.
Memory Layers + Cairn System
/build-agent-conversations-cairn
Reads your recent Claude conversations. You approve the L1 waypoints. L1 + L2 + L3 land in your vault.
- 7DObsidian: a visual graph your agents read and write.
- 8DPostgres + pgvector: real semantic search at scale.
- 9DGraphRAG: entity + relation graph, answers that trace their path.
- 10-12DNeo4j: a production knowledge graph, multi-hop reasoning.
Explore the Stack.
More than an email digest.
The morning brief is not the finish line. It is the first tile of your Agent OS Dashboard - one surface where your whole fleet reports in.
Email, calendar, your top priorities. Pulled together and delivered before you sit down.
Every agent, every source, one pane. What changed, what needs you, what the fleet already handled.
Tyler's command center: a morning brief that grew into a full operations board. You will see it later.
LUNCH
Back in 20 minutes.
The Morning
Brief.
You built it all day. Now watch it snap together.
Every block you built today snaps into the brief.
Ship the Morning Brief, live.
Your turn: run it.
/build-morning-brief gmail,calendar (+ ~/briefing-config.md PRD)
What comes
next?
You shipped one brief. Here is the ceiling it climbs toward.
Your brief becomes a command center.
Not a feed you skim. A surface that shows what changed, what needs a decision, and what your fleet already handled. Tyler's Pepper board is the live example.
Email, calendar, Slack, iMessage, GitHub: every channel summarized and ranked before you open a single app.
Your Cairns feed it directly. It already knows what you decided last week and who you owe a reply.
What ran overnight, what worked, what needs a human call. One click to approve, delegate, or dig in.
The morning brief is one expression.
The Agent Native OS is the full framework.
Point it at any business. The same five modules show up every time.
SUMMER
CAMP!
Take the whole stack all the way: ship production agents, build a real business on it.
You leave with a working Agent Native OS
for one real business.
Runs on as little as Claude Code. Scales to a full self-hosted fleet we share openly. The framework meets you at whatever level your stack is at.
Your visual map + teacher for the repo.
Everything today, and everything you did not get to, lives here. Your private guide to the agent-native-os repo: prompts, skills, the full run of show. Copy, paste, keep building.
