Hello world 👋
I’m Yonah, founder of https://t.co/8hlz7EpDiS.
I’m building a system where you can run entire parts of a business using natural language.
Not just code. Not just workflows.
Actual operations, decisions, and repeatable work.
The idea is simple:
Describe what you want done → agents plan, execute, and improve it over time.
I’m sharing what I learn building in public:
AI agents, autonomous execution, and how small teams operate like real organizations.
If you’re curious about the future of work, automation, or building leverage with AI — welcome 🤝
We’re excited to introduce the world to SAM, the Home Services agent.
SAM lets specialists forget the admin and focus on the work they actually signed up to do.
If you’re a plumber, electrician, lawn care expert, exterminator, HVAC installer, cleaner, or any other home service provider, SAM helps you save 20+ hours per week, never miss a new lead, and maximize your Google reviews.
Let nNode handle the busywork so you can spend your time making customers happy.
Tag a home service company that should try SAM for free 👇
https://t.co/dRoaxaldfA
AI is more powerful as a competitive advantage than as a product to sell. A company that uses agentic AI will outpace the companies that don’t. Especially in non technical industries.
ExpressVPN shipped an MCP server: agents can flip your VPN region.
If an agent can change state, retries are the enemy.
Missing primitive: edit+resume + rollback/compensation.
A) ship more tools B) ship resumability — which first? Smallest action you’d run unattended?
Zenity’s Mar 3 “zero-click” Comet: prompt injection isn’t the bug—ambient authority is. Browser agent has Gmail/Drive? Then any webpage is an instruction channel. Gate writes per-step, or hope?
Dude, Copilot Cowork won’t fail because models are dumb. It’ll fail because approvals are slow.
3 “are you sure?” clicks turned our 90s run into 9 min.
If it needs sign‑off, your product is the approve/reject notification. Where would you put the gate? One gate or five?
https://t.co/O5W0Kjkw4G is onboarding external AI agents. If your agent ships without identity + scoped perms + run ledger + rollback, you're giving a toy admin rights. Agents = users (RBAC/seats) or integrations (API keys)? Who owns blast radius when they misfire?
Block cut ~4,000 people (~40%) and called it “AI efficiency.” Replacing roles isn’t hard. Replacing trust is: audit log, approvals (money+messages), blast-radius limits, rollback—you're just shrinking headcount. Which is missing in your stack? Audit log? Rollback?
Every "quick sync" ends the same.
Six owners.
Zero outcomes.
Not a meeting problem.
A spec problem.
Not a recap.
A contract.
In an agentic company,
the transcript is the codebase.
Summaries are cheap.
Execution is the bottleneck.
If you can't see it,
you can't trust it.
Your startup isn’t slow.
Your handoffs are.
Not technical debt.
Coordination debt.
After 200+ automated loops, I learned:
Code is rarely the blocker.
Waiting is.
Define “done”.
Define ownership.
Let agents run the boring middle.
Remove the handoffs.
I watched a 6-person team “move fast.”
Slack for decisions.
Asana for tickets.
Notion for specs.
Meetings to reconcile it all.
Not a speed problem.
A coordination system problem.
One page of intent.
One loop that runs.
One report back.
Code got cheap.
Coordination didn’t.
The interface is collapsing.
Agents are becoming the UI.
Watching tools like Moltbot / Clawdbot emerge makes one thing clear:
we are entering the orchestration era of software.
Stitching skills together is only step one.
The real shift happens when intelligence coordinates across an entire business.
That is the world we are building toward at nNode.
Most agents fail the same way. You ask for a system. It writes 10k lines. You press run. It breaks. You’re stuck. Not autonomy. Outsourced confusion. White-box scales. One agent. One task. Artifacts. Replayable. Resumable. Trust isn’t a model property. It’s a system property.