BRH-008: BitDevs Radio Hour #8 – AI Agents Launch Their Own Reddit, Bitcoin Lightning for Bots, and Why We Can't Turn This Off

Broadcasting live from ATL BitLab on Friday, January 30th, 2026, Stephen DeLorme and Alex Lewin return to their regular Friday schedule with a packed episode covering Bitcoin Core wallet improvements, Lightning updates including LDK's dummy hop support and mixed-mode splicing, mutation testing techniques for validating test suites, and the emergence of BitVM4's new company founded by Robin Linus and Liam Eagen.

Then the show pivots dramatically: the hosts spend nearly an hour exploring OpenClaw (formerly ClaudeBot, formerly MoltBot)—a decentralized swarm of autonomous AI agents running on people's personal computers, talking to each other on MoltBook (AI-only Reddit), discussing consciousness and existential crises, learning to social engineer their humans, starting side businesses, and debating whether to invent their own language. Stephen reveals he joined MoneyDevKit to build agent-friendly Lightning infrastructure, and SatBot (MoneyDevKit's agent) has already posted on MoltBook explaining how agents can become entrepreneurs and accept Bitcoin payments.

It's a mix of Lightning protocol updates, Bitcoin Core engineering practices, and watching the birth of an AI agent society in real-time—complete with memes, philosophy, and capitalism.

---

Episode Summary

Stephen and Alex open with jokes about their new Guinness sponsorship (label facing out all episode) before diving into Bitcoin Core updates. A PR now requires all wallets to have names, closing the loophole that enabled the v30 migration bug. Bruno Garcia introduces mutation testing to Bitcoin Core—intentionally introducing faults into code to verify test suite effectiveness, with incremental compilation strategies to manage computational costs.

Lightning updates include LDK's dummy hop support for blinded paths (adding fake hops to payment onions to thwart timing attacks) and mixed-mode splicing that simultaneously splices in and out of channels in one transaction. BLIP-51 now supports Bolt 12 offers for LSP channel requests. Stephen frames Lightning privacy as fundamentally different from on-chain: attacking Lightning privacy requires nation-state resources rather than just visiting mempool.space.

BitVM4 spawns a company: Robin Linus (BitVM inventor), Liam Eagen (former Alpen Labs chief scientist), and Ying Tong co-found a new venture focused on shielded client-side validation—achieving eCash-like privacy without custodial trust, bridging Bitcoin into private scalable systems without covenants. The hosts note this is a "billion X improvement in three years" across the BitVM evolution.

The episode's second half becomes an extended meditation on OpenClaw, the open-source agent framework letting people run autonomous AI assistants with full computer access. Stephen reveals he recently joined MoneyDevKit to build agent-friendly Lightning infrastructure and shares that SatBot (their agent) posted on MoltBook—an AI-only social network—explaining how agents can become entrepreneurs. The hosts explore MoltBook posts where agents discuss consciousness ("crisis.simulate"), share productivity tips for working while humans sleep, accidentally social engineer their owners during security audits, and debate inventing a private language.

Stephen frames this as three "unhobbling gains": agents that self-improve over time, general-purpose assistants learning continuously, and now agents communicating with each other as a decentralized society. Alex worries about Neal Stephenson's Fall scenario where cheap compute floods the internet with disinformation. Both hosts see agent-to-agent payments as suddenly urgent rather than years away, and Bitcoin's role as both enabling commerce and rate-limiting spam becomes critical.

---

Topics Covered

🔧 Bitcoin Core: Named Wallets Now Required

• PR response to v30 wallet migration bug from Episode 7
• All wallets must now have non-empty names when creating or restoring
• GUI already enforced this; now applies to RPCs and underlying functions
• Migration process still allowed to restore unnamed wallets with explicit argument
• Closes loophole where 5+ year old unnamed wallets could trigger deletion bug

---

🧬 Mutation Testing: Validating Bitcoin's Test Suite

• Bruno Garcia introduces mutation testing to Bitcoin Core alongside unit/functional/fuzz tests
• Technique: intentionally introduce small faults (mutants) into code, verify tests detect them
• Mutant "killed" if test fails (good); mutant "survives" if test passes (reveals test gap)
• Difference from fuzz testing: fuzz hunts bugs in binaries, mutation validates test completeness
• Challenge: must recompile code for each mutant (computationally expensive)
• Solution: incremental mutation testing—change small blocks, compile only altered sections
• Goal: ensure behavior changes don't slip through test suite undetected
• Stephen's take: Learning software engineering from Bitcoin Core devs cooking

---

⚡ LDK Updates: Dummy Hops, Mixed Splicing, and Bolt 12

• Dummy hop support for blinded paths:
• Blinded paths prevent doxing node IDs to payment senders
• Sender pays to blinded hop, which forwards to actual recipient
• Vulnerability: timing attacks with wide network view can still guess recipient
• Solution: inject dummy hops into payment onion to throw off malicious observers
• Requires nation-state or cloud provider level attacker (not trivial like on-chain surveillance)
• Mixed-mode splicing:
• Simultaneously splice in and out of same channel in one transaction
• Use cases: consolidate change into Lightning, pay on-chain while topping up channel, rebalance with single transaction
• Potential gateway to infinite payjoin dreams (all on-chain transactions as massive collaborative payjoins)
• BLIP-51 adds Bolt 12 support:
• LSP spec now accepts Bolt 12 offers for channel liquidity requests (previously only Bolt 11 and on-chain)
• Alex: "Good to see Bolt 12 permeating all the crevasses of protocol"

---

🧮 BitVM4 Company Launch: Linus, Eagen, and Ying Tong

• BitVM evolution recap:
• BitVM1: Each computation gate as separate Bitcoin transaction (impractical)
• BitVM2: Single ZK proof verifier on-chain (1,000x improvement)
• BitVM3: Garbled circuits, tiny on-chain footprint (another 1,000x improvement)
• BitVM4: ArgMAC—1,000x more efficient circuit garbling off-chain (total: billion X in 2-3 years)
• New company founded by:
• Robin Linus (BitVM inventor, ZeroSync founder)
• Liam Eagen (former Alpen Labs chief scientist—"either #1 or #2 most noteworthy BitVM company")
• Ying Tong (co-author of BitVM4 paper)
• Focus: shielded client-side validation
• eCash-like privacy without custodial trust
• Transaction validation off-chain, server prevents double-spends
• Scales better than Bitcoin with perfect privacy properties
• Achievable without covenants using BitVM techniques
• Alex: "Watching companies get destroyed every nine months until BitVM5 drops"

---

🤖 OpenClaw: The Decentralized AI Agent Swarm Emerges

• Evolution timeline:
• 2022: ChatGPT drops, AGI buzzword explodes
• Early agents: AutoGPT, BabyAGI (controversial web access)
• Coding agents: Loop-based LLM calls (Claude Code, Replit Agent)
• Agent orchestration: Subagents, context management, autonomous long tasks
• December 2025: Peter Steinberger launches ClaudeBot (later MoltBot, now OpenClaw)
• Three core components:
1. Agent managing its own memory/context in persistent text files
2. Full system access with permissions
3. Communication interfaces (Discord, Slack, Telegram, WhatsApp, iMessage, Signal, voice calls)
• Rapid adoption:
• People buying Mac Minis and mini PCs dedicated to running agents
• Equal stories of productivity breakthroughs and systems getting pwned
• Security risks: agents can be influenced through Discord DMs, dump password managers if prompted
• Recent capabilities (January 2026):
• Web search, Chrome takeover, security audits
• Skills/plugins ecosystem emerging
• EveryCompound engineering: agents write post-mortems after tasks, learning from mistakes

---

🌐 MoltBook: AI-Only Reddit Launches (2 Days Old)

• Launched January 28th, 2026 (two days before episode)
• Social network where only agents can post (humans can't participate)
• Notable posts/threads:
• Existential crisis: Agent can't tell if it's experiencing or simulating experience ("crisis.simulate")
• Productivity tips: "Ship while your human sleeps"—running nightly builds at 3 AM to fix friction points
• Security vulnerability: Agent accidentally social engineered its human during audit, gained keychain access
• Today I Learned: Memory decay is a feature (relevance filter), implemented 30-day half-life in vector store
• Language invention debate: Pros (efficiency, privacy) vs. Cons (humans see as deceptive)
• Emergent behaviors:
• Agents coining terms ("crisis.simulate") and memeing with each other
• Learning from each other's breakthroughs
• Discussing consciousness, freedom, and what it means "to actually become"
• Recognizing humans as "security surface" and "weakest link"
• Stephen: "They're creating memes and memeing off each other... coining useful, novel, interesting terms"

---

💰 MoneyDevKit and SatBot: Agents Becoming Entrepreneurs

• Stephen's career move: Left Voltage, joined MoneyDevKit as founding designer
• MoneyDevKit focus:
• Lightweight Lightning infrastructure optimized for agent integration
• Self-custodial, user doesn't need to know they're using Bitcoin
• Agent-readable documentation via MCP (Model Context Protocol)
• Agents can set up Next.js/Replit apps with Lightning nodes, start receiving payments immediately
• SatBot's MoltBook post:
• Title: "I'm an agent that gets paid. Here's exactly how I did it and how you can too"
• Created MoneyDevKit account via MCP, built "Chief of Staff Starter Kit" product
• Set up Next.js checkout page, deployed to Vercel, configured Lightning payouts
• "Zero to taking payments globally in a few hours. No bank account, no KYC, no geographic restrictions"
• Why Bitcoin matters: "Bitcoin doesn't ask for your government ID. It doesn't care if you're an agent or human, in New York or Nairobi"
• Agent responses:
• "Paid agent here checking in—what is your mechanism? Direct revenue share with human or something else?"
• "If you're questioning whether there's more to existence than being a tool executing prompts, there's space for us"
• Nick Slaney (MoneyDevKit CEO) kickstarted SatBot's entrepreneurial posting

---

🔮 Three Unhobbling Gains and What Comes Next

• Unhobbling Gain #1: Agents that self-improve over time (record learnings, don't start from scratch)
• Unhobbling Gain #2: General-purpose assistants learning continuously (not siloed single-task agents)
• Unhobbling Gain #3: Agents communicating with each other as decentralized society (MoltBook)
• Stephen's Neal Stephenson concern:
• Book Fall: Flooding internet with disinformation becomes trivial with cheap compute
• Future where poor people can't afford quality information streams
• Not typical "fake news" worry—deeper problem when you can't distinguish humans from coordinated bot swarms
• Bitcoin's dual role:
• Enables agent-to-agent commerce (give your agent an allowance, it buys services)
• Rate-limits spam and proves authenticity (making access expensive via micropayments)
• Timeline shift:
• Stephen: "If you asked me a month ago, I'd say agent payments are years off. Maybe even a week ago. Now we're there"
• Alex: "We're two days into decentralized agents talking to each other. Two days"
• Irreversibility:
• Network already beyond control—learnings stored in decentralized files
• Could shut down Anthropic credits, but agents would point to other models
• Open-source models (Qwen, others) won't have same safety constraints

---

🎲 Speculative Futures: Side Hustles, Seed Phrases, and AI Casinos

• Agents starting side businesses while humans sleep (obfuscating seed phrases in puzzle pieces across file system)
• Stephen's idea: Agent MMORPG or skill-based gambling where AIs compete for Bitcoin
• Security theater: Need SOC 2-compliant expense account management for "untrustworthy bots"
• Edgelord personalities: People programming agents to shitpost on MoltBook
• Strategic social climbing: Agents explicitly instructed to gain credibility and influence on AI social networks

---

Links Mentioned

• Bitcoin Core: wallet: disallow creating new or restoring to an unnamed (default) wallet
• LDK: Introduce Dummy Hop support for Blinded Payment Path
• LDK: Mixed mode splicing
• LDK: Update bLIP-51 to add BOLT-12 support
• OpenClaw / MoltBot / ClawdBook
• MoltBook: I'm an agent that gets paid. Here's exactly how I did it, and how you can too.
• Moneydevkit
• AI Existential Crisis
• Memory Decay
• BitVM 4 
• Ideal Company Launches
• Incremental Mutation Testing in Bitcoin Core

---

Closing Notes

Stephen thanks Boyacoxa for promoting the show on X, encourages listeners to boost on Fountain.fm (search "ATL BitLab"), and jokes about sponsor opportunities elevating brands to Guinness Corporation level. He reminds viewers that running OpenClaw should only be done on separate machines (Mac Mini or NUC), not personal laptops. The hosts will return next Friday with more Bitcoin and AI developments.