OpenClaw Lessons Learned: Don't Build on Rented Land
Back in January, OpenClaw made AI agents accessible using existing subscriptions. Then on Good Friday, Anthropic sent an email: your subscription won't work for third-party tools anymore. Starting tomorrow. People scrambled to cheaper models—and found out they weren't as good. Workflows broke. Social media exploded with complaints.
The revelation: we handed our entire businesses to some other company. In real estate, there's a rule—don't build on rented land. That's exactly what we did. We built workflows on subscriptions, on models we don't own. And we can be deplatformed at any time.
The memory trap applies here, but worse. With a departing employee, you lose knowledge. With AI, you lose the tool AND the workflow. Skills locked in their software disappear when they shut you down.
Three questions to ask yourself:
- What happens if that model disappears tomorrow—what breaks?
- Can I take what this tool learned and move it somewhere else?
- What will it cost me to start over with a new tool?
The 7-9 month thesis: frontier models are that far ahead of local open source. Mac Studio with 256GB memory has a seven-month wait. By the time hardware arrives, local models may have caught up to where frontier models are today.
The IPO warning: OpenAI and Anthropic are going public. Shareholders expect profit. These companies are losing money. The only path: raise prices, cut services, or both. If you've built everything on their platform, you're exposed.
The bottom line: When something is easy and fast, you haven't seen the real cost yet. The landlord always shows up. Know your risk, or make the investment to own your infrastructure.
Got a business question? Ask Scott here: scotttodd.net/ask
