About Paul Lopez
Forward-deployed AI architect. More than 35 years in technology, from mainframes to agentic AI.
What I do now
I'm part of Accenture's Advanced Technology Center, embedded directly with client teams across insurance, retail, financial services, and healthcare. The job is to take an AI initiative from an open question to an agentic system running in production, on the client's own data and cloud.
My work centers on the gap between a promising model and a system a business can actually run: the context it draws on, the boundaries around what it's trusted to decide, the abstraction that keeps it portable across cloud providers, and the instrumentation that catches trouble before a customer does.
What I wrote in 2018
“Deep Learning will have more impact on humanity than the Internet.”
That year I opened my blog with a bigger claim: that AI is the inflection point of all inflection points, one that would change how we think not only about computers and machines but about what it means to be human. In 2018 it read as a bold bet. Deep learning was still mostly prediction and classification, and most enterprises were still debating whether to try it at all.
Since then the bet has paid off faster than I expected. Large language models moved AI from recognizing patterns to generating and reasoning, and agents now write code, read contracts, and take actions through tools. The questions I raised back then about ethics, transparency, bias, and control are no longer research topics. They are design decisions every enterprise makes when it ships an agent: what should this system be allowed to do, and who decides?
Seventy years in the making
None of this started recently. In 1957 Frank Rosenblatt, a project engineer at the Cornell Aeronautical Laboratory in Buffalo, New York, built the Perceptron, the first single-layer neural network and a founding moment for cognitive science. It took decades of cheaper compute and far more data for that idea to reach everyone with basic programming skills, and then everyone with a keyboard.
The path here
My career has followed each platform shift: networks, the internet, cloud, data, and now AI. I've worked across healthcare IT, financial services, telecommunications, semiconductors, and enterprise software, building systems people actually use, from telecom networks serving millions to healthcare platforms processing billions in claims. That's why I believe the most useful AI comes from deep domain knowledge combined with hands-on building, not from theory alone.
- AccentureAdvanced Technology Center: forward-deployed AI architect, agentic systems
- Optum (UnitedHealth Group)Principal Healthcare AI Architect, Optum.ai
- NEC Corporation of AmericaVice President, IT Services & Operations
- T-Mobile USA, Nortel Networks, IBM, AMD, National SemiconductorKey roles in engineering, product strategy and market development
I hold a B.S. in Electrical Engineering from the University of Texas at Austin and completed MBA coursework at SMU's Cox School of Business and the Wharton School. As an IEEE member, I contributed to standards committees for computer networking and wireless technologies.
Writing since 2014
Every post from my earlier blog now lives here with its original date. A few from each era:
2014–2015
Crowds, startups and developers
Crowdfunding, crowdtesting and the developer economy, written while the cloud was turning every garage into a software company.
2017–2019
Containers, blockchain and data
The infrastructure years: containers as the new unit of work, trust in enterprise contracts, and data platforms before models got big.
2020
Deep learning gets real
Deep learning leaves the lab: medical imaging, fairness in training data, and why so many AI projects failed to reach production.
2023–2024
Generative AI arrives
Large language models, mixture of experts, and the first agents that could use tools instead of only talking.
2025–today
Agents in production
What it takes for agentic systems to survive real data, real users and real audits: protocols, evaluation, trust boundaries and cost.