All posts
Autonomous agents in real operations: what we learned in production
Six months running agents with human approval in back-office processes. What worked, what failed, and why the human boundary matters more than the model.
How we evaluate RAG systems before putting them in production
A three-layer evaluation framework: retrieval, generation, and behavior in the real business workflow.
Multi-agent orchestration: when it makes sense and when it does not
More agents is not better. Coordination patterns that reduce latency and errors in complex flows.
Automation with approvals: the model that scales best
Not everything should be 100% autonomous. The sweet spot is automating 80% and reserving the critical 20% for humans.
LLMs in LATAM: infrastructure, latency, and real costs
Latency and cost-per-token benchmarks for deployments in Mexico and Colombia. Field data, not marketing.
Conversational memory in long-running agents
How we design memory that persists across sessions without leaking sensitive data.
Observability for agents: traces, costs, and decisions
What to log when an agent makes autonomous decisions and how to audit without slowing the system.
Fine-tuning vs prompting: when to invest in each
Practical rules from 12 projects. 90% of cases do not need fine-tuning.