I wrote a framework called 5 Filters for Deciding Under Uncertainty. Then I realized: I should eat my own cooking.
Over the past two weeks, I've been building an agentic AI system using OpenClaw on a Mac mini. I'm not a software engineer. Multiple YouTube videos warned me not to do this unless I was a "hardcore network admin with full confidence in my own coding capabilities."
I did it anyway. Let me apply the 5 Filters to the decisions I've made.
The Context
I work at a large company with software engineers. My manager recently recommended we read a textbook on agentic design patterns. Instead of just reading, I decided to build.
Participatory learning beats theoretical study — but was that the right call?
Decision 1: Buy the Mac Mini
Filter 1 — Optionality: ✅ HIGH
Creates dedicated AI experimentation environment. Portable, low-cost entry point (~$600). Can repurpose or resell if needed.
Filter 2 — Reversibility: ✅ EASILY REVERSIBLE
Can sell the Mac mini (high resale value). Total sunk cost: ~$600 + time. No contractual obligations.
Filter 3 — Information Value: ✅ HIGH
Could NOT have learned this from reading. Real constraints (ports, memory, processing) vs theoretical knowledge.
Filter 4 — Premortem
- Mac mini underpowered → Start small, scale if needed
- Can't figure out setup → Engineering colleagues, community support
- Project abandoned → $600 is course tuition; acceptable loss
Filter 5 — 10-10-10
- 10 minutes: Excitement, some anxiety ("Can I actually do this?")
- 10 months: Either working system OR learning experience that informed understanding
- 10 years: Understood AI agents by building them. Career differentiation.
✅ STRONG DECISION
Decision 2: Use OpenClaw (Unverified, Open Source)
Filter 1 — Optionality: ✅ HIGH
Open source = no vendor lock-in. Can modify, extend, fork. Community vs. corporate dependency.
Filter 2 — Reversibility: ✅ REVERSIBLE (with caveats)
Can uninstall, switch platforms. Time invested is real sunk cost. Data portability: building in standard formats.
Filter 3 — Information Value: ✅ VERY HIGH
Textbook: "Agents can be orchestrated." Building: "Orchestration fails when context windows overflow."
Filter 4 — Premortem
- Security vulnerabilities → Regular backups, isolated environment
- Community disappears → Open source = have the code
- Months wasted on setup → Time-box, get to "hello world" quickly
✅ STRONG DECISION
Decision 3: Build Instead of Read
Filter 1 — Optionality: ✅ HIGHEST
Reading gives knowledge. Building gives capability. Capability > Knowledge in uncertainty.
Filter 2 — Reversibility: ✅ FULLY REVERSIBLE
Time spent building is "wasted" only if you learn nothing. You always learn something when building.
Filter 3 — Information Value: ✅ MAXIMUM
Textbook: "Agents can be orchestrated." Building: "Orchestration fails when context windows overflow."
Filter 5 — 10-10-10
- 10 minutes: "Should I just read the book?"
- 10 months: I've built something; colleagues are still planning to read the book
- 10 years: I have intuition; they have theory. Both matter, but intuition is rarer.
✅ OPTIMAL DECISION
Decisions Not Yet Made
| Open Question | Key Filter | Suggested Action |
|---|---|---|
| How much access to grant the system? | Reversibility | Start scoped, expand gradually |
| Share with engineering team? | Information Value | Share selectively with one trusted engineer |
| How much time to invest? | 10-10-10 | Set explicit boundaries |
| Productize or keep personal? | Optionality | Defer. Use personally for 3+ months first. |
The Honest Assessment
When I looked at this analysis, my first reaction was: "This feels generous."
I haven't done a formal premortem. I'm two weeks in. The current state is messier than this framework suggests.
But here's what the framework revealed: I'm more confident about the 10-year possibility than the 10-minute one. The 10-month horizon is the sweet spot — concrete enough to plan, far enough to show real results.
The Meta-Point
The 5 Filters assume you're making decisions under uncertainty. That's exactly what I'm doing with agentic AI — nobody knows how this evolves.
In uncertainty:
- Optionality > Optimization (building capability, not optimizing known processes)
- Reversibility matters (low commitment, high learning)
- Information is valuable (building reveals what reading can't)
- 10-year thinking wins (early in a transformative domain)
My manager's team is reading textbooks. In 10 months, I'll have built something. In 10 years, I'll have intuition they can't match.
The framework validates my approach: keep building, stay scoped, defer productization, and trust that participatory learning outperforms theoretical study in uncertainty.
The fish are reading about the ocean. I'm building a submarine. 🌊