Port33’s open-source AgentOS seeks to connect engineering activity, human and machine contributions, and the business results used to justify AI spending.

The uncomfortable question about enterprise artificial intelligence is no longer whether employees are using it. It is whether anyone can explain what that usage produced.

Uber offered a particularly visible example in 2026. The company reportedly exhausted its annual budget for AI coding tools within four months, while its president and chief operating officer acknowledged that it remained difficult to connect rising Claude Code consumption with a corresponding increase in useful customer-facing features. The tools were being used. The financial return was harder to trace.

For CFOs, allocators, and technology leaders, that distinction sits at the center of AI ROI. A usage dashboard can show how many tokens a team consumed or how frequently employees opened an assistant. It cannot necessarily tell you which combination of human judgment, agent activity, and engineering work created a measurable result.

From Usage Metrics to Attribution

Port33 AgentOS approaches the problem as one of attribution and coordination.

The open-source system is designed to record work through what Port33 calls a contribution ledger. According to the company, the ledger tracks activity across human developers and AI agents, assigns contributions, and creates a clearer record of how a project moved from its stated goal to its completed outcome.

That framing moves AgentOS beyond the category of a conventional developer utility. For a financial audience, its intended role is closer to that of an enterprise AI accountability layer, one that provides a record that decision-makers can examine when evaluating costs, ownership, and results.

The thesis is straightforward: weak returns may not always indicate a weak model. They may reflect fragmented objectives, duplicated effort, unclear responsibility, or an inability to connect engineering activity with business value. AI engineering coordination therefore becomes part of the capital-efficiency question.

A Governance Problem Viewed Through Capital

Afsheen Afshar brings an institutional perspective to that argument. Before founding Pilot Wave Holdings, he served as chief AI officer at Cerberus Capital Management, following earlier roles as JPMorgan’s first chief data science officer and a managing director at Goldman Sachs.

His experience deploying AI across a private equity portfolio shaped his view that technical activity requires governance at the point of execution. A board deciding whether to expand an AI program needs more than evidence that employees adopted the software. It needs to understand where the spending went, who or what contributed to the output, and whether the result supports further AI capital allocation.

Pilot Wave provided the operating context for AgentOS. The firm applies AI within acquired industrial companies, giving its teams direct exposure to the practical problems of deploying systems in real workflows rather than evaluating AI solely as a standalone investment theme.

Connecting Engineering Work With Investment Decisions

Calvin Cooper, a director at Pilot Wave and co-founder and chief operating officer of Neurometric, adds a complementary perspective. His background includes venture capital, public markets, company building, and advising at the intersection of finance and technology.

Through Neurometric, Cooper is also exploring questions of model cost and task-specific performance. That work extends the accountability thesis into another environment: organizations need to understand not only whether an AI system performs, but whether its cost and output make sense for the job assigned to it.

For fintech AI leaders, the relevance is immediate. When you assess an AI program, aggregate usage says little about whether capital was deployed efficiently. Attribution offers a more useful starting point.

Open Source as a Form of Scrutiny

Port33 released AgentOS as an open-source project in July 2026. Its founders describe that choice as central to adoption, allowing engineering teams and enterprise stakeholders to inspect the system, test its assumptions, and contribute to its development.

That openness does not establish adoption or prove financial outcomes on its own. It does, however, allow the proposed governance mechanism to be evaluated rather than treated as a closed claim.

The broader idea behind Port33 AgentOS is that enterprise AI spending requires an auditable connection between work and value. As board-level questions become more specific, companies with credible answers may be better positioned to defend their budgets than those that still present activity as achievement.