Senior AI & full-stack engineer
Eric AguayoCode with Intent. Build to Endure.

Engineer. Builder. Trail runner.
Hey there, I’m Eric Aguayo!
I’m a senior AI and full-stack engineer based in Quito, Ecuador, and a trail runner at heart. I’m Director of Applied AI at Nimblersoft. I build AI products and help engineering teams bring coding agents into their development workflows. I design and lead agentic development workflows with critical plan review, tests, explicit completion criteria, and traceability between requirements and changes.
I’m available for part-time or per-project contract work, and for mentoring teams and engineers on AI agents. Let’s talk about your project.
Selected projects:
- Vespera: an app to help plan your day and protect your focus, with a suggested agenda and research-informed micro-actions.
- Agent Fabric: open-source tool to install and sync shared agent definitions across OpenCode, Kilo, Codex, Claude Code and more, without overwriting your customizations.
- Artifacts Manager: view and manage agent-generated HTML and Markdown artifacts across projects.
Lately I’ve been writing about what really costs in an agent workflow: not the tokens, but the attention (in Spanish).
This blog is where I share my journey, from tech experiments to trail adventures and everything in between.
Latest Posts

What No One Understands, No One Maintains (Part 1): A Proposal to Evaluate Human Understanding
Orchestration matters, but I think human understanding matters even more: without it, logic errors slip past unnoticed, the system becomes hard to maintain, and scaling can turn into chaos. My proposal: have AI ask us questions before closing a task.
If You Don't Practice Them, They Fade: What Urku Trail Reminded Me About AI
I ran the 10K at Urku Runas Trail in La Rinconada: 10 km, around 1,000 m of elevation gain, in about two hours, not in my best shape. The cobblestones remind you that the fundamentals (your ankles, and your judgment too) need ongoing training. The same happens when we offload too much cognitive work to AI.
A Mistake in Your Prompt Can Lead to Unwanted Results
A mistake in a prompt can produce a coherent yet unwanted result. A reflection on natural language, programming, and the need to validate intent.
The Expensive Part of an Agent Isn't Tokens, It's Attention
An agent can cost pennies in model usage and still eat up your week. In my workflow, the expensive part is the attention it demands: steering, plan review, fixing drift. Here is how I try to cut that down before the agent writes the first line.
Stop Reading LLM Text Walls: Why I Built a Local Artifacts Manager for Agentic Workflows
Reviewing thousands of lines of LLM markdown is a massive cognitive bottleneck. Here is why I moved beyond disposable chatbot canvases and built a persistent, local-first Artifacts Manager for AI agent planning and architecture.