automation3.0

$ a3 create morning-brief.a3

Welcome to Automation 3.0

Where natural language becomes the automation layer

Write what should happen. Agents do the work.

Intent → Trigger → Agent → Outcome

ABOUT AUTOMATION 3.0

English is no longer just the instruction. It is the automation.

Automation 3.0 is the idea that natural-language instructions can become durable, executable operations. Instead of drawing a workflow, you describe the outcome, timing, tools, and boundaries. AI agents interpret the intent and carry the work forward.

Natural language automation

Write the operation in plain English—the same language used to explain the work to a capable teammate.

Agent-native execution

Hermes, OpenClaw, Claude, Codex, or another agent becomes the runtime that reasons and acts.

Universal triggers

Run on a schedule, when an event happens, when data changes, or whenever a person asks.

Human-defined boundaries

Prompts carry permissions, approval gates, stop conditions, and the judgment that must stay human.

ALIGNED WITH SOFTWARE 3.0

From English as code to English as operations.

Software 3.0 asks what happens when natural language becomes a programming language. Automation 3.0 extends that shift into ongoing work: a prompt can define what runs, when it runs, which agent acts, and where a human steps in.

Explore the Software 3.0 concept

SOFTWARE 3.0Prompt→Implementation

Natural language becomes working software.

AUTOMATION 3.0Prompt→Continuous operation

Natural language becomes work that can keep running.

THE PARADIGM SHIFT

The interface keeps moving closer to intent.

Every generation removes another layer between a person and the outcome. Automation 3.0 replaces the visual wiring diagram with an instruction that agents can understand and execute.

Automation 1.0

Scripts & cron

Machines follow exact procedures.

Hard-coded logic
Fixed schedules
Maintained by engineers

Automation 2.0

Visual workflow builders

People connect every step.

Drag-and-drop nodes
Manual branching
Canvas complexity

Automation 3.0

Prompt-native operations

Agents pursue the stated outcome.

Natural-language intent
Adaptive execution
Human control by policy

$ outcome becomes operation_

AUTOMATION 3.0 IN ACTION

One instruction. Three readable views.

The prompt stays the source of truth. The execution plan makes agent behavior inspectable, and the run log shows what actually happened.

Automation3 Enhanced — morning-brief.a3

# Weekday Support Brief

Every weekday at 8:30 AM, ask Claude to summarize
new support tickets. Let Codex prepare draft fixes.

Requirements:
- flag anything urgent
- do not change production
- ask for approval before sending the brief

trigger: schedule(weekdays, 08:30)
agents: [claude, codex]

steps:
  1. read(new_support_tickets)
  2. claude.summarize(priority = true)
  3. codex.draft_fixes(write = false)
  4. human.approve(before = send)

status: ready

08:30:00  trigger fired · weekday schedule
08:30:02  Claude read 24 new tickets
08:30:08  summary complete · 3 urgent
08:30:11  Codex prepared 2 draft fixes
08:30:12  production writes blocked by policy

08:30:14  waiting for human approval_

Automation3 EnhancedLn 9, Col 1UTF-8A3

The prompt is the durable operation.

Toggle between human instructions and machine execution
Run on schedule, by event, or on demand
Choose the best agent for each kind of work
Keep approvals and boundaries beside the intent

AGENTS ARE THE RUNTIME

Hermes AgentOpenClawClaudeCodexAny capable agent

FREE THAI EBOOK · 27 CHAPTERS

Go deeper into the agent-first era.

Business OS for the Agent-first Era turns the Automation 3.0 concept into a practical operating model: goal, context, tools, approval, proof, and memory working as one system.

Full interactive readerSearchable chaptersPresentation mode

Read the Ebook View presentation

Cover of Business OS for the Agent-first Era

Read free in Thai

THE AUTOMATION 3.0 MANIFESTO

Automate the outcome, not the diagram.

A useful concept needs principles. These four keep prompt-native automation understandable, portable, and under human control.

01
Intent is the source
The human-readable instruction remains authoritative. Generated plans are implementations of that intent.
02
Agents are the runtime
The operation names the outcome and boundaries; capable agents decide how to complete the work.
03
Triggers are context
Time, events, data changes, and human requests decide when the instruction becomes active.
04
Humans define the limits
Permissions, approvals, budgets, and stop conditions travel with every automation.

$ imagine the next operation.a3

What if every recurring process began as a sentence?

Automation 3.0 is a concept for the agent-first era. Explore the curated open-source ecosystem, challenge the idea, and build on it.

Browse open-source projects Read the manifesto