AGI Leash Length

Measure AI capability by how long you can let it work autonomously before intervention

Dan Shipper
The AI-native startup: 5 products, 7-figure revenue, 100% AI-written code.

AGI Leash Length

"You can tell how much better AI is getting by how long a leash you can give it to do work... I think a good definition of AGI is when does it become economically profitable for people to run agents indefinitely?" - Dan Shipper

What It Is

AGI Leash Length is a practical framework for thinking about AI capability and progress toward AGI (Artificial General Intelligence). Instead of abstract benchmarks or Turing tests, it measures AI advancement by a simple, observable metric: how long can you let it work autonomously before needing to intervene?

The framework draws an explicit parallel to child development (via psychologist Winnicott): just as children develop by being gradually given longer periods of autonomy—first needing constant attention, then supervised play, then unsupervised time, then full independence—AI systems are progressing along a similar arc.

Under this model, AGI is reached when it becomes economically profitable to run agents indefinitely—never turning them off, never waiting for your next instruction. They're off "living their lives" productively like a trusted employee.

How It Works

The progression of AI autonomy:

Generation Autonomy Duration Example
Autocomplete Seconds Tab completion in Copilot
Single-turn chat ~1 minute ChatGPT Q&A
Multi-turn conversation 5-10 minutes Extended Claude conversation
Agentic tools 20-30 minutes Claude Code, Deep Research
Future AGI Indefinite Always-on productive agents

The economic test for AGI:

The key insight is adding "economically profitable" to the definition. You could technically run any agent in a loop forever, but that doesn't mean it's AGI. True AGI is when:

  1. The agent can work indefinitely without intervention
  2. The value it creates exceeds the cost of running it
  3. It's doing useful work proactively, not just waiting for instructions

How to Apply It

  1. Benchmark your tools - For any AI tool you use, ask: "How long can I let this run before I need to check on it?" This tells you its practical capability level.

  2. Match tool to task - If a task requires 30 minutes of autonomous work, use a tool that can handle that leash length (like Claude Code). Don't expect a chat interface to handle complex multi-step work.

  3. Track the trajectory - Watch how leash length changes with new model releases. Claude Opus 4 can work autonomously longer than earlier versions. This is the real capability metric.

  4. Plan for the future - As leash length increases, the tasks you can delegate will expand. Prepare your workflows for a world where you delegate full projects, not just tasks.

  5. Know current limits - We're at about 20-30 minutes of autonomous work today. That's roughly "toddler in the same room" level—you can step back, but you're not leaving for the day.

When to Use It

  • When evaluating which AI tool to use for a task
  • When setting expectations for AI autonomy in workflows
  • When explaining AI progress to non-technical stakeholders
  • When planning how AI will change your work over time
  • When cutting through AGI hype to focus on practical capabilities

Source

  • Guest: Dan Shipper
  • Episode: "The AI-native startup: 5 products, 7-figure revenue, 100% AI-written code."
  • Key Discussion: (00:15:09) - Dan presents the leash length framework and AGI definition
  • YouTube: Watch on YouTube

Related Frameworks

Back to Decision Making