---
title: CEO Defends Governance of AI Agents
url: https://www.elseif.net/ceo-defends-governance-of-ai-agents
published: 2026-10-19T17:03:59+00:00
language: en
section: Models
source: https://itforum.com.br/noticias/fabiana-falcone-governanca-ia-agente/
organizations: Synapser, IT Forum, McKinsey, Gartner, Brasscom
publisher: elseif
---

# CEO Defends Governance of AI Agents

CEO Fabiana Falcone defends governance of AI agents at IT Forum Upload event. She argues AI agents require distinct governance from traditional software. Falcone states technology errors are reproducible and governed as assets. People cost payroll and follow four pillars: authority, budget, evaluation, termination. AI interprets, chooses, and costs per use. She notes AI errors differ from software errors. Same input can yield different outputs with confidence and conviction. Errors may not repeat, complicating correction. Falcone applies four pillars to agents. Authority requires no hiring without approved position, yet any employee can create agents instantly. She cites a sales executive who built a CRM agent causing unallocated costs. Budget involves token costs charged after usage. Evaluation measures AI by token savings. Synapser reduced consumption to 10% of prior levels with new architecture. The key question is whether customer service remained unchanged. Termination lacks clear criteria for replacing agents. Responsibility remains human. Agents decide on behalf of companies. Who placed the agent, defined limits, and should monitor? Falcone states this issue moved from CIO to CEO. It involves cost, revenue, and fiduciary responsibility.

She presents data on AI adoption challenges. IT Forum Intelligence reports 68% of CIOs cite AI adoption as top challenge by 2026. McKinsey research shows only 6% of companies attribute about 5% of results to AI, a figure stagnant for two years. Gartner projects token programming costs will exceed average developer salary by 2028. A major bank already spends more on tokens than payroll. Token costs split between human resources and IT without either seeing total. Falcone notes code generation is no longer the bottleneck. Governance of generated output becomes the new constraint. Business users prototype, test, and present ready projects to IT, increasing demand queues. She describes a credit superintendent who solved a critical issue with prompt programming, labeled as an infringer when IT discovered unauthorized AI use, or shadow AI. Other CIOs created governed pipelines for such programming. Security risks include agents reading emails and acting on hidden phishing instructions, potentially leaking sales pipelines. Falcone stresses the need to define minimum agent authority after email reading. She cites a technology talent deficit of 797,000 professionals reported by Brasscom, with only 53,000 graduates annually. She questions the source of senior professionals as the pyramid base shrinks.

She advocates for generalists who know a little about many areas. Reskilling for judgment layers like security, testing, and disaster recovery is essential. She states people with fear do not think, they execute, requiring professional involvement in process redesign. Pressure comes from all sides. SaaS startups face service-as-software where intelligence replaces code, allowing internal development. Technology service firms like Synapser see clients demanding this model, forcing internal process reviews. Traditional software partners feel revenue pressure as clients avoid renewing licenses. McKinsey reports companies capturing value redesigned workflows with human-agent logic, not applying AI to existing processes. This represents an operational model decision, not just IT, affecting measurement and compensation. Measuring staff by logged hours prevents automation efforts. Many pilots fail to reach production for this reason. Falcone emphasizes governance frameworks must address these challenges. The industry must shift from code-centric to governance-centric approaches. AI agents demand new accountability structures beyond technical implementation.