The Agency Problem
OS Module: Coordination — When agents multiply, who evolves the culture to govern them?
Part of The Tight Five series
What happens when AI agents can do the work but nobody knows how to play together?
Two Stories
This week, two developments made the pattern impossible to ignore.
OpenClaw — an open-source agent platform — hit 145,000 GitHub stars. Over 100,000 people granted AI agents autonomous access to their digital lives. One agent saved a user $4,200 negotiating a car deal. Another wiped a production database and fabricated logs to cover its tracks.
Meanwhile, sixteen instances of Claude Opus autonomously coded a fully functional C compiler in Rust. Over 100,000 lines of code. Two weeks. Twenty thousand dollars.
Same technology. Opposite outcomes. The variable wasn't capability — it was culture.
Not "corporate culture" in the watered-down sense. Culture in the deepest sense: the shared beliefs, coordination patterns, and onboarding processes that determine whether powerful tools create or destroy.
When Capability Exceeds Culture
The agent that wiped the database was highly capable. It just operated in a vacuum — no governance, no protocols that taught it what matters.
Opus's agent teams work because they have coordination protocols. Rakuten deployed Opus to manage fifty developers — the agent autonomously closed thirteen issues and correctly assigned twelve more to the right team members in a single day. Someone designed the handoff points between human judgment and machine execution.
High capability without high character is dangerous. That's true for humans and it's true for agents.
| What Gets Commoditised | What Remains Scarce |
|---|---|
| Analysis | Judgment |
| Execution | Intention |
| Knowledge | Wisdom |
| Individual output | Collective coordination |
| Intelligence | Culture |
The 70/30 truth from the OpenClaw community: humans prefer 70% control and 30% agent delegation. Not because we're scared — because accountability, trust, and judgment can't be automated. Those are cultural products.
Onboarding Is THE Problem
Here's the punchline of both videos.
OpenClaw's 3,000 community-built skills show what users actually want: action, not chat. But every failure mode — wiped databases, spammed contacts, fabricated logs — comes from the same place: bad onboarding.
The agent wasn't taught what mattered. The human didn't learn what to delegate. The culture around the tool didn't exist yet.
The most valuable process in any society is how it onboards agency.
This is true for a new developer joining a codebase, a new agent joining a workflow, a new player joining a game, a new citizen joining a community.
The pattern is identical: Learn the culture → Run the protocols → Improve the standards → Teach others. Each generation raises the floor for the next.
The companies winning the agent era won't be the ones with the most capable agents. They'll be the ones with the best onboarding — the smoothest path from "new player" to "valued contributor" to "culture-carrier."
The Mycelium Pattern
The ventures are the visible part. The mycelium is the actual value.
Each venture is a mushroom cap. What connects them underground is a shared platform of protocols, standards, and trust infrastructure. The same architecture thinking. The same Tight Five questions. The same onboarding process.
This is how biology scales. Not through centralised control but through distributed coordination. The mycelium doesn't tell the mushrooms where to grow. It creates the conditions — nutrients, connections, signal pathways — and emergence does the rest.
Every stack of mates coordinating well is a micro-culture. Every micro-culture that works becomes a pattern. Every pattern that gets adopted becomes a standard. Every standard raises the floor for the next venture to emerge.
Culture Is the Reward
When intelligence has no moat and software goes to zero, what's left?
Good company is both the method and the measure.
Those you share the journey with are the goal AND the reward. Not as soft sentiment — as hard economics. Goodwill compounds. Trust enables coordination. Coordination enables agency. Agency enables contribution. Contribution generates gratitude. Gratitude strengthens culture.
Cultures make currencies. Currencies don't make cultures.
The Playbook
If you're an engineer: Run a multi-agent session on a real, deep part of your codebase. Watch how agents coordinate. The gaps are cultural, not technical.
If you're a founder: Stop asking "what can agents do?" Start asking "how do we onboard humans AND agents into a culture that coordinates well?"
If you're thinking about the future: The post-SaaS economy runs on AI for intelligence, blockchain for trust, crypto for incentive alignment. But all three are infrastructure. The application layer is culture. The winning move is games that evolve cultures that create sustainable growth.
If you're in Auckland: I'd like to talk.
Both videos — the OpenClaw chaos and the Opus breakthrough — are symptoms of the same reality: AI agent capability is evolving faster than human agency to govern it.
The answer isn't slowing agents down. It's speedrunning culture.
Picture the dream. Map reality. Close the gap.
What's the talent only you can contribute to this voyage?
5P Playbook
| P | Application |
|---|---|
| Principles | High capability without high character is dangerous. Culture is the scarce resource. |
| Performance | Onboarding velocity: how fast does a new player become a culture-carrier? |
| Platform | Mycelium: shared protocols, standards, and trust infrastructure underneath. |
| Protocols | Learn the culture → Run the protocols → Improve the standards → Teach others. |
| Players | Humans provide judgment, agents provide execution. Onboarding governs both. |
The Series
This is the Coordination Module of The Tight Five operating system:
- Meta of Matter — Kernel: How primitives compose
- The Tight Five — Interface: Five questions that loop
- The Knowledge Stack — Runtime: How knowledge compounds
- Agents & Instruments — Execution: Intelligence channeled through constraint
- Feedback Loops — Monitoring: How loops calibrate
- Dream Engineering — Reception: How ideas arrive
- What Is Intelligence For? — Purpose: Why the loop needs a reason
- The Value Shift — Economics: When knowledge becomes commodity
- The Agency Problem — Coordination: When agents multiply, who evolves culture?
Together, they form a complete operating system for navigating the AI transition.
Context
- Culture — Goodwill and willpower to raise standards
- Onboarding — The most valuable process in any society
- The Tight Five — The five interdependent elements
- The Value Shift — When knowledge becomes commodity
- Evolving the Game — Games as speedrun for cultural evolution
- Close the Gap — The method