The bottleneck is not the model. It's the harness.
Harness Engineering: the emerging discipline of building what surrounds the model so that agents reach production with reliability and governance.
- Head of AI · Google Cloud LATAM
- Ex-NTT DATA
- Agentic AI
- Conversational AI
- Robotics
- Product
- Enterprise transformation
EUREKA, every two weeks in your inbox
AI, agents and robotics — ideas, talks and analysis for going from demo to real-world impact. In English and Spanish.
You will confirm on Substack. No spam; unsubscribe anytime.
Five theses I'm betting on
Convictions that shape how I read what's happening in AI — and where the real opportunity sits.
- 01
The bottleneck is the harness, not the model.
Real performance of an AI system depends on what surrounds the model — tools, context, validation, memory, limits — more than on the model itself.
- 02
From copilot to Digital Employee.
Autonomous agents don't assist anymore: they execute complete processes end to end. That changes how we measure value, risk and ROI.
- 03
Robotics just found its brain.
Foundation models are what robotics waited forty years for. Embodied AI is the next frontier — industrial, scientific and social.
- 04
Cost per outcome will decide which agents survive.
Tokens, retries, human oversight: every agent has a P&L. The advantage is not having agents — it is knowing which ones pay their own way, and being able to prove it.
- 05
If it doesn't reach people, it's still a demo.
Applied AI that matters = business + society + people. Without all three, there's no real impact.
Where to start
A handful of pieces that capture the current thinking — agentic AI, foundation models and the realities of shipping them.
The agentic enterprise needs an immune system
How to design AI agents that can act without concentrating trust: identity, least privilege, containment, oversight, and recovery.
The Stopwatch and the Exam
Static AI benchmarks did not die; they stopped being enough. A grounded look at the move from capability to agency, the harness problem, and where to put attention now. Backed by a public catalog of 69 agentic benchmarks.
Cognitive Architecture and Emergent Phenomena in Advanced AI
A synthesis of current mechanistic and behavioral research on how cognitive structures and emergent behavior arise in modern AI systems.
The next generation of AI: Self-Improvement and Autonomous Learning
Self-improvement and autonomous learning in AI — and the road to an intelligence explosion.
What I write about
Recurring themes across the library. Filter the library by any topic.
Prototypes, benchmarks and tools
Interactive experiments to explore AI, agents and complex systems — starting with an open AI-agent benchmark tracker.
A probabilistic map of where the episodes of the Iliad and the Odyssey might have happened — each location classed as accepted, plausible, speculative or mythical, scored 0–12 against a published rubric, with sources, rival theories and reusable JSON/GeoJSON.
Interactive sandbox: configure an agent on six axes, pick an OWASP LLM Top 10 threat, switch the eighteen controls of the Agentic Control Matrix on and off, and read the coverage — quadrants filled, quadrants left empty, frameworks answered. Deliberately without a residual-risk score.
An interactive taxonomy that encodes 24 AI agents — from OpenAI, Anthropic, Google, Microsoft, Amazon and SpaceXAI to the open-source frontier — as vectors across six orthogonal axes plus an action surface, with a filterable table, an A×T×I governance-risk matrix and a machine-readable JSON API.
Recommendations
A handful of voices from teams I've built, executives I've worked with and engineers I've shipped with.
“During the time we worked together he demonstrated a remarkable vision to apply technology in order to solve real daily problems. His passion and proactivity makes him achieve what sometimes seems impossible. Very focused on customer needs, always a pleasure to work with Santiago!”
Agustin Aznar GabásHead of Retail (Spain, Portugal & Italy) · Executive Committee“Santi worked with me for 10 years. His ability to set up teams and his foresight in detecting new tech trends, that will disrupt the market is outstanding. He ran the product management of a set of payment products, which are successfully used today by Spanish Merchants, Banks, and Users.”
José Manuel ReyesPayment Processing Specialist“Santi is constantly surprising with innovative and disruptive products with huge potential to change the market. He is always ahead of the pack in terms of technology and how it can disrupt businesses.”
Angel Luis de la MoyaProduct Manager · Getnet Europe (a Santander Company)“I worked on Santi's team for 4 years. He has a surprising drive, he is able to break down all the barriers to launch a product. Due to his software engineering background he understands the technical challenges, which makes communication between clients and technical staff easy.”
Héctor Abraham Morillo PrietoSoftware Engineer · Automattic
Where to find me
Same ideas, different formats. Pick the channel that fits how you read, watch or listen.
Who's behind this
Physicist by training, today Head of AI Forward Deployed Engineers for LATAM at Google Cloud. I write to make sense of the AI shift — and to share what works when you actually have to deliver it.