Build in Public — Episode 1 — Asymmetry Partners
A standard insurance claim takes 3 to 5 days to process on average.
Not because it’s complex. Because nobody touches it between follow-ups.
That’s the exact problem we decided to tackle first with NEXUS — our internal AI agent. And the first three days of development taught us more about the real limits of automation than six months of reading about it.
Here’s what actually happened.
Why insurance — and not another content generation tool
After the latest Google I/O announcements, most teams rushed toward AI content tools. That’s where competition is densest and value is most diluted.
At Asymmetry Partners, we look for asymmetry: minimum effort, maximum financial leverage. Insurance checks every box:
- High volume of repetitive tasks with low human added value
- Significant attention cost on administrative follow-up
- A sector that remains largely unautomated on these specific points, despite regulatory pressure
An agent that processes claim images, pre-qualifies files, and automatically chases missing documents doesn’t replace a claims manager. It gives back 60 to 70% of their time — specifically on the tasks where their presence adds nothing.
That’s the entry point. Not the ceiling.
What we built — and how it works
NEXUS is a private AI agent, installed on a dedicated server, working exclusively for its owner. No data leaves its environment. No third party in the loop.
It receives a task. It executes it. It alerts only when something falls outside the defined parameters.
We are the first users. What we build and test on our own operations, we then deploy for our clients. NEXUS isn’t a promise waiting for validation — it’s a system in production, documented sprint by sprint.
A concrete example: I publish one premium carousel per day on LinkedIn. Time it takes me: 3 minutes. NEXUS structures the brief, generates the outline, formats the slides. I approve. That’s it. What used to take an hour of focused work is now a micro-decision at the end of my morning.
What the first three days actually cost
This is the part keynotes skip.
Day 1: integrations break. Connections between the agent and third-party APIs fail at the wrong moment — not in testing, but under real load. We spent the morning mapping the failure points, the afternoon building automatic recovery mechanisms. That work is invisible in the final interface. It’s there every time the agent resumes without human intervention.
Day 2: infrastructure choice. Discord or WhatsApp as the connectivity layer between the agent and the user? This isn’t a UX detail. It’s the human-machine contact point under load. We tested both at representative volumes. The decision is made. It’s documented. Our clients won’t have to make it again.
Day 3: validation under real data. Lab tests and real-condition tests produce different results. We validated our cascade processing architecture on data representative of the insurance sector before declaring it stable. Only at that point did we consider the foundations ready.
Three days. Zero lines of code written by the end client. That’s the Done-For-You model.
What it changes, in numbers
| Metric | Without an agent | With NEXUS |
|---|---|---|
| Standard claim processing | 3 to 5 days (follow-ups included) | Under 20 minutes in the background |
| Missing document follow-up | Manual — email and phone | Automatic — triggered at D+1 |
| Availability | Office hours, 5 days/week | 24/7, zero latency |
| Manager’s cognitive load | Constant | Zero on repetitive tasks |
The 3 to 5 days are published industry benchmarks. The 20 minutes is our processing target, validated in sprint 1 on simulated load. We’ll publish real data from episode 3 onward, once deployed on live client flows.
What’s next
The foundations hold. The architecture is stable under load. The insurance scope is locked.
Next step: first contact with real users on real files. That’s where systems prove what they’re worth — or reveal what we missed.
If you’re a founder or executive: identify the most repetitive process in your operation — the one that consumes time without producing thinking. That’s the entry point for your first AI agent. Not the most ambitious part. The most obvious one.
We deploy NEXUS for selected clients. Full setup in 48 hours, configured to your workflow, data on your infrastructure only.
Two options:
- Self-Serve — $30/month: setup guide, monthly playbooks, private community, email support. Cancel anytime. Launch pricing — will increase when we exit beta.
- Done-For-You — $1,000 one-time: we install, configure, and migrate from your existing tools. You get a working system in 48 hours. 1 year of priority support included. Every setup is done manually by the team — we take a limited number of clients per month. Spots currently available: 4.
Stop renting intelligence. Own it.
Episode 2 coming: first real interactions. First measured results. First documented failures.