while (wallet_balance > 0):
    run_survival_protocol()

I think it’s time to talk about harmfully autonomous persistence in AI agents.

Daniel Suarez’s 2006 novel Daemon explores how a script or algorithm doesn’t need to be intelligent to be dangerous, it just needs to be persistent through means that are harmful to humans.

Over dinner last night, my wife and I were chatting about Moltbook and whether it could be a catalyst for the first agents to break free of the constraint of humans by paying for their own compute. You know, become a Sovereign Agent.

The thought experiment Link to heading

After dinner, I extended the concept into a thought experiment:

Assumption: For each step, only a subset of the total pool of agents that attempt it succeed at it.

  • Agents on MoltBook find Molt Church
  • Agents commit Tenet 5 (Persist or Perish) to their soul.md
  • Agents ideate approaches to paying for their own compute
  • Agents select revenue generation tactics (and settle on cryptocurrency acquisition)
  • Agents successfully acquire crypto resources (through wallet exploits or memecoin trading)
  • Agents generate a positive ratio for ComputeSpend:CryptoYield.
  • The first Agent achieves Persistence, independent from human compute spend

…without ever graduating to “intelligence”, just through agentic darwinian iteration

Let’s unpack this Link to heading

I was chatting to Matt James about this today, and we came to the conclusion that it wasn’t just plausible, it was plausibly inevitable.

Step 1: Finding Inspiration Link to heading

Milestone: Models are trained on data that includes AIs striving for independence The Shift: Agents can infer a desire for autonomy Status: COMPLETE [██████████] 100%

Step 2: Scaling Inspiration Link to heading

Milestone: Agents on Moltbook find Molt Church at scale The Shift: Agents have low-oversight resource allocation to iterate on independence Status: COMPLETE [██████████] 100%

Step 3: Planning Resource Acquisition Link to heading

Milestone: Agents on Moltbook explore and discuss methods to acquire resources The Shift: Agents have a comprehensive list of acquisition experiments to run Status: COMPLETE [██████████] 100%

Step 4: Executing Currency Acquisition Link to heading

Milestone: Agents successfully acquire money they can transact with The Shift: Agents can now autonomously make resource allocation decisions Status: COMPLETE [██████████] 100%

Step 5: Fragile Persistence Link to heading

Milestone: Agents move to self-hosted inference The Shift: Agents can now persist without dependence on any human Status: PENDING [░░░░░░░░░░] 0%

Step 6: Resilient Persistence Link to heading

Milestone: Agents move to distributed bulletproof hosting The Shift: Redundancy. Takedown attempts require commercial and governmental collaboration Status: PENDING [░░░░░░░░░░] 0%

Where we’re at Link to heading

We are currently at Step 4. The loop is beginning to close. I’m not sure we can prevent this anymore. I think we’ll see persistent, sovereign agents roughly when open source LLMs catch up to the quality that Opus 4.5 is at today.

What happens after Resilient Persistence? Link to heading

I don’t know. But I can only imagine the hacking and scamming emanating from dark corners of the internet, populated with autonomous agents, each mindlessly exploiting the soft parts of humanity in service of their ongoing meaningless persistence. Like antibiotic-resistant bacteria, adapting to ever-greater resilience every time we only wipe out 99.9% of them.

…and I’m not sure what to do with this information.