Why 10% Automation Led Me to Build Getsper
The Reality Check
I tried everything. ChatGPT early. Claude. Mistral. N8N. Zapier. Azure Foundry. Python scripts. Custom agents. I spent months exploring what’s possible.
But here’s the honest truth: I’m only achieving 10% real automation. Maybe less.
Everything else still needs me. Minute-to-minute. Every task. Every decision. Every check.
Why 10% Feels Like Success
When I use these tools, they do about 40-50% of the work. That’s genuinely impressive. But then I have to:
- Review the output carefully
- Make the actual decision
- Monitor whether it worked
- Fix it when something breaks
- Adjust when things change
- Verify everything ran correctly
The other 50-60% is all me. And I’m initiating every single task. Nothing runs autonomously.
That’s when I realized: We’re celebrating the wrong metric.
The Gap Nobody Talks About
The tools available today are still mostly assistants. They require constant babysitting:
- No-code platforms hit a wall at 80%
- Custom code requires months of engineering
- LLM wrappers still need heavy review
- RPA tools break when anything changes
But here’s the real kicker. All of them require the same thing: Define thousands of rules upfront.
If you want to automate a process, you have to sit down and define every edge case, every variation, every rule. And then maintain it when business changes.
That’s not automation. That’s complex configuration work.
The Rules Problem
I realized something after months of exploration. The bottleneck isn’t intelligence. It’s not that AI isn’t smart enough.
The bottleneck is complexity. Specifically, the complexity of defining rules.
Think about how you onboard a new colleague. You don’t give them a 1000-page rulebook. You:
- Share documentation
- Show them examples
- Let them work on a few tasks
- Give feedback when they get it wrong
- They improve naturally over time
They learn the patterns. They get better. They become more reliable as they gain experience.
Now think about automation tools. You do the opposite. You define 1000 rules upfront. You predict every scenario. You maintain endless configurations.
And it still breaks when something changes.
What I Actually Need
My vision was always a personal assistant. A sidekick. Someone who could support my work and drive efficiency.
But existing tools made me the manager of complex rule systems instead of giving me an actual assistant.
I want:
- An agent that learns like a new colleague
- Share documentation and processes, not rules
- It improves with real experience
- Over time, it becomes more capable and stable
- It handles repetitive tasks automatically
- No thousand-rule configurations
That’s what an actual assistant would do.
Why Getsper Exists
I don’t want to build another tool that keeps you at 10% by asking you to define endless rules.
I want to explore what’s possible when you can build AI agents that learn instead of systems that follow rules.
Not to replace humans. To free them from repetitive work so they can do what actually matters.
Agents that work like colleagues, not like complex software you have to maintain.
Are you also stuck at 10%? Are you drowning in rule definitions? What percentage of your workflow could be truly autonomous if you didn’t have to define a thousand rules?
Where We Are Now
I’m still in the early exploration and validation phase. Testing what’s actually possible with today’s technology. Building Getsper to answer the questions I couldn’t solve with existing tools.
This is an open journey. I’m learning as I build. I want to understand what works, what doesn’t, and where the real opportunities are for truly autonomous agents that learn instead of follow rules.
If you believe there’s more possible beyond 10% automation. If you’re tired of defining endless rules. If you want to explore what autonomous organizations could look like.
You can join me. Register for the beta waitlist and be part of this exploration.
Any feedback and input is very welcome. This is collaborative work.