.png){kind=logo}
The self-taught engineer from Uruguay built a career across food, education, pandemic health software, financial-grade identity, and now the emerging field of AI agent authentication.
Before Paola Estefania de Campos De Franco began building identity systems for AI agents, she ran a vegan restaurant. Before she worked on financial-grade identity at Auth0, now Okta, she was teaching food and wellness courses across countries. Before she created the Agent Auth Protocol, she had already left technology once, rebuilt her career, and learned that systems reveal themselves under pressure.
“My path does not look traditional,” de Campos says. “But that is exactly why I arrived here.”
de Campos is now a Staff Software Engineer at Better Auth, where her work sits inside one of software’s newest security questions: how should systems handle AI agents that act on behalf of people without pretending to be those people?
Her route to that question began in Montevideo, Uruguay. She started studying computer science at a private university in 2012 and worked at a newspaper, but she did not finish the degree. In 2015, she made a sharp turn away from software and into food. She trained in the United States as a vegan and health coach, founded AnnaPurna, a vegan food business in Uruguay, and later ran her own 100 percent vegan, organic restaurant.
That chapter may seem far removed from software security. de Campos does not see it that way.
“Running a restaurant teaches you how systems fail when pressure is real,” she says. “You learn very quickly that a process that works on paper can break when people, timing, inventory, and expectations all collide.”
She also worked as a chef on a yacht in the Bahamas and taught courses across Uruguay, Argentina, Paraguay, Panama, Colombia, and India. In India, she taught for free as a service. The work required her to teach, organize, adapt, and manage people across different contexts.
The return to software came through a practical need. In 2019, de Campos built her own online-courses website. The project pulled her back toward engineering, not as an abstract interest, but as a tool she needed to make something work.
“I did not return to software through a perfect plan,” she says. “I returned because I needed to build something, and the more I built, the more I remembered why engineering had always attracted me.”
Then COVID arrived. de Campos returned to engineering full-time during the pandemic and built hospital registration software for Doctors Without Borders, also known as MSF. That work placed software inside an urgent setting, where usability, reliability, and speed mattered.
“Health software during a pandemic is not a place where you can treat engineering abstractly,” she says. “People need the system to work because the situation around it is already difficult enough.”
From there, her software career accelerated. She moved into senior and lead engineering roles in Spain, then joined Auth0, now part of Okta. At Auth0, she worked on financial-grade identity for regulated industries and authentication for AI agents. The experience introduced her to the identity systems that quietly decide who can access what, under which rules, and with which proof.
That work changed her understanding of trust.
“Identity systems are invisible when they work,” de Campos says. “But the digital world depends on them. They decide whether access is safe, whether actions can be traced, and whether people can trust the systems they use.”
At Auth0, she contributed to the async authorization features (CIBA) and to auth for agents, work that involved async auth and token exchange as a whole, including DPoP and another spec. She also saw the limits of trying to adapt identity systems built around human users to AI agents that can act independently. The more agents began doing things for people, the more uncomfortable the default model became.
An agent using a user’s account may look convenient, but it creates a confused record. If the agent sends an email, reads a file, or changes data, the system may say the user did it. That creates a problem for security, auditability, and control.
“The idea of something that is not you doing things as if it were you gave me the creeps from the beginning,” she says. “It felt like the industry was normalizing the wrong behavior before we had named the risk.”
That discomfort became the problem she chose to solve. Now at Better Auth, an open-source authentication framework downloaded more than 2 million times each week, de Campos works on the Agent Auth Protocol, bringing her unconventional path into one of software’s newest trust problems.
In simple terms, the work is about giving agents a safer way to act for people without erasing the difference between the user and the software.
“We do not need agents to borrow human identity,” she says. “We need systems that make the delegation clear.”
That distinction reflects both her technical experience and her non-linear career. de Campos has had to prove herself through visible work rather than conventional credentials. She is a woman engineer from a small South American country, without a completed university degree, competing in one of software’s most technical fields: identity and security.
“The unconventional path is not something I try to hide,” she says. “I have had to prove myself with shipped systems, not credentials.”
Open source became part of that proof. Better Auth’s reach gives her work exposure inside a large developer ecosystem, and the Agent Auth Protocol is now being tried by early partner companies. She has also participated in San Francisco developer events with Better Auth alongside companies such as Vercel, Strapi, and Mux.
Still, de Campos' focus remains practical. She is not interested in agent identity as a theoretical category. She is interested in what happens when people begin trusting agents with real actions.
If an agent can book travel, send email, access money, update systems, or handle company data, people need a way to use it without surrendering control of the accounts those agents touch. To de Campos, that is not a luxury feature. It is the foundation of trust.
“People should be able to use powerful agents without losing control,” she says. “That is the line I care about.”
Her goal is to help make agent authentication part of the early infrastructure, not an afterthought. The work is technical, but the principle is human: people should be able to use digital assistants without giving up the boundaries that protect them.
That is the thread connecting her unusual path. A restaurant, a classroom, a hospital registration system, a financial-grade identity platform, and an open agent protocol all taught her something about systems under pressure.
“Trust is built before the crisis,” de Campos says. “Whether you are running a kitchen or building identity infrastructure, the system has to be ready before everything depends on it.”
For more information, visit the Agent Auth Protocol website.
