AI agents have become essential enterprise tools. The platforms companies choose now will decide whether they own their AI or depend on big tech. Mozilla CTO Raffi Krikorian argues that open-source AI agents are the key to real independence.Key points:AI agents act inside your systems, and most enterprises have limited visibility into whose interests they serveBuilding on closed, proprietary platforms means the vendor controls your agent's behavior, data access, and roadmapKrikorian explains the open-source alternative and what enterprise control of AI actually requires 

Agentic AI is changing what CIOs are accountable for. Systems that plan, act, and call tools on their own now operate within workflows the CIO no longer fully owns, while boards expect the same standards of security, risk management, and business value. CXOtalk episode 919 turns that pressure into a practical playbook for leading agentic AI in the enterprise.Key points:The CIO mandate shifts from running systems to governing autonomy, with clear decision rights, agent boundaries, and accountability.Trust, data, and control must be managed together, especially when the middle layer of models, agents, and vendors is opaque, and shadow AI is already inside the business.Human oversight must be designed for machine speed, with explicit roles before, during, and after AI operates, backed by an operating model and culture built for continuous change.

AI-driven autonomous software development is transitioning from pilot projects to real production deployments in 2026. CIOs managing large developer teams must decide which tasks to delegate to agentic platforms and the speed of this transition. CXOTalk episode 918 examines how Mexico's largest insurer made those decisions, including pilot design, measured velocity gains, changes to the developer role, and the governance inputs required to keep autonomous output aligned with enterprise compliance standards.Key PointsTest Autonomy Across Mixed Use Cases First. GNP structured their pilot around four concrete scenarios: backend language upgrade, frontend framework migration, new feature builds, and security vulnerability remediation, using live repository and CI/CD connections.Move Guardrails into the Prompt Layer. Treat technical standards, security policies, and test requirements as prompt inputs alongside functional specifications, so the platform produces code that meets corporate guidelines by design.Redefine Developer Roles Around Direction, Not Code. Shift engineers from line-by-line coding into prompt authorship, architecture review, and validation of autonomous output, with co-pilots handling any residual work.

In 2026, insurance technology leaders face an important question: can autonomous development meet the AI determinism, auditability, and quality standards that regulated industries require? This conversation examines how a multinational insurer rebuilds its software development life cycle around AI, covering context engineering, test-driven development, throughput optimization, and the shifting bottlenecks that surface as code generation accelerates.Key PointsRegulated Industries Require Deterministic Code from AI. Regulated insurers need deterministic, auditable code rather than probabilistic output, which shapes vendor choice, context engineering from codebases and standards, and test-driven development.Optimize Throughput, Not Local Efficiency. Accelerating one SDLC stage exposes new limiting factors downstream. Treat requirements, code generation, review, testing, and release as one integrated, measured pipeline rather than isolated wins.Instrument AI Spend Against Actual Business Outcomes. Track velocity, quality, and end-to-end throughput against AI investment, so spend ties to faster product delivery and customer value, not isolated gains.

Every enterprise software decision in 2026 now runs through the same question: what does agentic AI actually do to my stack, my costs, and my vendors? We explore how agents deliver production value, which SaaS categories are genuinely exposed, and ideas for CIOs on the coming shakeout.Key points in this episode:Develop Agent Governance Before You ScaleTreat agents as software with authority to act, but define data access, approval rights, evaluation tests, security limits, and step-by-step monitoring before broad deployment.Match Autonomy to the WorkStart with bounded workflows that have clear inputs, established rules, and measurable outputs, such as coding, customer support, financial reconciliation, or search, and retain human approval for regulated or ambiguous work.Manage AI Spending Based on Capacity and BudgetTrack tokens, API calls, and tool use against business outcomes; then fund agents based on priority and ROI rather than letting every team consume budget on open-ended experimentation.

Coordinated AI agent swarms can now fabricate grassroots consensus, infiltrate communities, and corrupt enterprise AI training data at scale. This episode examines a 22-author Science study that maps how these swarms operate and what organizations can do about them:How AI swarms manufacture synthetic consensus that manipulates public and corporate discourseWhy your AI training data is a target and what "LLM Grooming" means for model integrityThe governance frameworks, economic levers, and detection methods that raise the cost of manipulationKey PointsAI Swarms Manufacture Public Opinion at Scale. Autonomous AI agents coordinate across social platforms to generate posts, likes, and shares that no human observer can reliably distinguish from authentic activity. These swarms self-optimize in real time, testing messages and amplifying whichever proves most persuasive, creating a convincing illusion of majority consensus around any narrative.Defenses Lag Far Behind the Threat. Launching an AI swarm requires minimal technical skill and inexpensive computing power, yet no reliable method exists to detect coordinated swarm behavior. Social media platforms have little incentive to close this gap because synthetic engagement inflates the daily active user counts, which they report to advertisers and shareholders.Corporate Reputation Is a Direct Target. AI swarms go well beyond political influence. Competitors and bad actors use them to fabricate grassroots boycotts, manufacture product safety scares, and coordinate harassment campaigns against executives and board members. Leaders must verify whether a wave of online backlash reflects real public sentiment or orchestrated manipulation before altering corporate strategy.

Marie Myers, Executive Vice President and CFO of Hewlett Packard Enterprise, explains how she moved agentic AI from advisory analytics into live finance operations using HPE's internal platform, Alfred. Key Points:Redesign Workflows Before You Deploy AI Agents. Standardize and centralize core finance processes before adding agentic AI. Deploying agents into fragmented workflows leads to failed pilots, while fixing the work first encourages faster adoption and measurable returns.Change Management Determines Whether AI Succeeds or Fails. The human side of change is the most challenging aspect of enterprise AI. Develop strict quality standards to avoid dependence on AI outputs, and maintain a "human in the loop" requirement for every AI-driven decision.Expand Your AI ROI Framework Beyond Hard Savings. Leaders should consider both direct returns and indirect value factors, such as speed, accuracy, error reduction, and fraud prevention, when evaluating AI projects.

AI agents are multiplying faster than the governance frameworks meant to control them. Cvent CIO Pradeep Mannakkara and CISO Ben Mayrides discuss how enterprise leaders can govern AI agents using the AWARE framework from Glean's Work AI Institute.You will learn:Why traditional security architectures break down when applied to autonomous AI agentsHow CIOs and CISOs can align on agent risk without slowing innovationPractical steps for building agent governance today