Founded to make lessons last.
LonsLabs was started in Spain in 2025 by people who kept watching good fixes disappear.
Troubleshooting is not learning.
Teams that run AI agents in real workflows are very good at fixing today's problem. What they rarely get is a system that remembers the fix, tests it, and carries it into the next version. The knowledge leaks out through chat threads, handovers, and retros nobody re-reads.
We built LonsLabs as the layer that keeps it. Observe the failure, update the harness, deploy with evidence — and write the lesson into the exact prompt, policy, tool, or evaluation that must change.
How we build.
Evidence before change
A proposal without a trace is an opinion. We link every edit to what actually happened.
Ownership is explicit
Every artifact has someone accountable for it, and our tooling respects that instead of routing around it.
Ambiguity is a signal
When two people read a rule two ways, the rule is the bug. We fix the rule.
Done means measured
We do not consider an improvement real until production confirms it.
Working on agents that need to get better every week?
We would like to hear where they fail.
Contact LonsLabs