The paperwork tends to arrive late. A model is built in a notebook, trained on a dataset living somewhere else, tested with a separate library, discussed in chat, reviewed in a meeting, and promoted through a ticket. Weeks later, someone asks for the record. The model may be finished. Its story is scattered across half a dozen tools.
Cyril Brignone has spent much of his career noticing this kind of gap. The technology changes. The organizational problem does not. Complex work gets distributed; responsibility stays stubbornly central. Somebody still needs a coherent view.
At Vectice, the San Francisco company he co-founded with Gregory Haardt in 2020, that view is built for AI and machine-learning teams. Vectice captures metadata and lineage, generates model documentation, and organizes review around the evidence produced during development. The company calls the category regulatory MLOps. Strip away the label and the idea is plain: make the record while the work is happening.
“We’re not replacing your tool, we’re giving a global view of all of your assets across your data tools.”Cyril Brignone, on Vectice’s original product thesis
The same question, in different clothes
Brignone arrived in Silicon Valley from France in 2000 to work at Hewlett-Packard Laboratories. He was a computer scientist entering a world of sensor networks, RFID, and what researchers called smart environments. He later led HP’s Smart Environment Research Group. Its work contributed to HP’s RFID and Sentient Spaces programs, attempts to make physical spaces and objects legible to software.
This was before the Internet of Things became a standard slide in corporate strategy decks. The technical question was not merely whether an object could send a signal. It was whether a larger system could understand the object’s identity, location, permissions, and context. Brignone’s work in distributed sensor networks and IoT produced 23 patents, according to the announcement that later named him CEO of Arrayent.
Then he moved from the lab to startups. In 2007, Brignone co-founded Vator.tv, an online platform where emerging companies could present themselves to investors and a wider startup audience. As COO, he helped architect the platform and establish partnerships with AOL, Entrepreneur.com, Microsoft, Google, and others. Vator looks different from a sensor network, but its product logic rhymes: gather distributed actors, standardize how they present information, and make a crowded system navigable.
From products to relationships
Brignone joined Arrayent in 2011 and worked across field operations, Europe, product, sales, and delivery before becoming CEO and a board member in 2015. Arrayent provided a cloud platform for brands that wanted to connect consumer products without building all the underlying infrastructure themselves. By the time Prodea acquired the company in 2017, Arrayent said its platform managed more than 1.5 million connected products across 82 countries.
His language from that period is revealing. In a 2016 interview, Brignone argued that connected products required manufacturers to shift from a product mentality to a service mentality. A sale was no longer the end of the relationship. Connectivity created ongoing obligations: uptime, software updates, maintenance, support, and a two-way flow of information.
He named the two big obstacles “transformation” and “commitment.” The technology forced companies to rethink who owned the customer relationship and what happened after an object left the factory. It also made interoperability essential. Products had to work with other products, clouds, and services if connected experiences were going to become useful rather than merely novel.
“I believe the future of connected products will be based on seamless interoperability.”Brignone, 2016
Those lessons travel neatly into AI. A model is also not a one-time artifact. It changes, depends on other systems, and creates obligations after deployment. It needs monitoring, maintenance, review, and an intelligible history. The technical work may be distributed across teams, but the institution remains accountable for the result.
Build beside the workflow
Vectice began with a broad knowledge problem. Enterprise data science teams were producing valuable assets, but the context around those assets was fragmented. Brignone described the company’s ambition as becoming a system of record without becoming another execution environment. Teams already had enough places to write code, train models, and store data. Vectice would sit across those places and provide the global view.
That constraint shaped the product. If documentation depends on a developer remembering to reconstruct months of work, it will always compete with shipping. If metadata and lineage can be captured as the work moves through existing tools, the record becomes a by-product. Reviewers can follow a model dependency map, inspect versions, and connect a claim in a document to the artifact underneath it.
The stealable idea
When experts resist a workflow, look for the missing by-product. Do not demand that they become enthusiastic archivists. Design the record to accumulate around the work they already value.
The need became sharper as Vectice focused on regulated financial institutions. Banks do not only need models that perform. Model developers, independent validators, risk teams, auditors, and regulators each need to understand how a model was built and challenged. Static documents age quickly. Evidence lineage answers a harder question: which exact dataset, code version, validation result, and approval supports this exact model?
Vectice has continued to turn that question into integrations and workflow. In 2025, it announced work with IBM watsonx.governance, introduced automated compliance checks, and announced an integration with ServiceNow. The compliance feature scores a model against a chosen policy set and links results back to versioned evidence. The aim is to make a review repeatable rather than a scavenger hunt.
The Nantes test
One story from Vectice’s early years says as much about Brignone’s operating style as any product diagram. In 2020, he and Haardt decided to establish an R&D center in France. They considered several cities, including Paris, and chose Nantes. Then the pandemic made travel impossible.
They proceeded from the United States. The founders hired a team, found office space, opened a bank account, recruited a technical lead, and learned the mechanics of French employment and accounting. Brignone later said they opened the office without having set foot in France. A company that took three days to establish in the United States required more than three months of work in France.
The telling detail is his appreciation for the local advisors who answered questions within 24 hours. There is no lone-founder theater in the account. The work succeeded through a network: regional agencies, an accountant familiar with foreign companies, a part-time CFO, and a technical lead willing to join remotely. On the founders’ first visit in September 2021, they finally met the team they had already assembled.
Brignone also said that every Vectice employee was a shareholder because “the value of the company is the value of its employees.” It is an operational statement dressed as a cultural one. Ownership, like documentation, makes responsibility explicit.
The infrastructure of accountability
Brignone’s career does not move in a straight line from one fashionable market to the next. It circles a durable systems problem. Sensors need context. Startups need visibility. Connected products need interoperability. Models need lineage. In every case, the activity is easy to create and hard to coordinate.
The Vectice bet is that AI governance should live close to engineering work rather than descend as a separate ritual. That position becomes more consequential as models enter regulated decisions and as generative systems add new components, prompts, agents, and dependencies. Speed produces more versions. More versions create more questions. Accountability requires a memory that can keep up.
There is a modesty to this kind of infrastructure. When it works, the developer keeps using familiar tools. The validator finds the right evidence. The risk team sees the state of the portfolio. The organization can explain what changed without staging an archaeological dig. The missing layer becomes visible mainly in its absence.
Brignone has been building versions of that layer for more than two decades. The objects have changed from tags and appliances to datasets and models. The ambition has stayed recognizable: make a complicated system understandable to the people who have to live with it.