Twenty questions before an AI agent reaches production
Production agents are not prompts with permissions. They are distributed systems with identities, tools, state, policy, cost and failure modes.
Read analysis →Lexdatum / Intelligence
Lexdatum Intelligence
Signals, architectures and operating decisions for the agentic enterprise.
19 analysis · July–August 2026 · EN + ES
Featured
Archive
- Analysis31 Aug 2026InteroperabilityMulti-vendor agent orchestration reaches mainstream enterprise softwareSalesforce’s Winter ’27 release brings A2A-based third-party orchestration into a major enterprise suite, validating interoperability as a procurement requirement.
- Analysis31 Aug 2026AI GovernanceThe agent registry is the new control inventoryAs agent estates scale, organisations need a governed catalogue of agents, tools, skills and MCP servers with ownership, approval, versioning and discoverability.
- Analysis29 Aug 2026SecurityThe Hugging Face incident: seven controls agent systems need nowOpenAI’s report on an internal evaluation incident is a warning against treating sandboxes, networks and shutdown paths as implementation details.
- Analysis27 Aug 2026InteroperabilityA2A becomes the horizontal layer of agent interoperabilityA2A’s move into the Agentic AI Foundation clarifies a two-protocol architecture: MCP connects agents to tools; A2A coordinates agents across boundaries.
- Analysis27 Aug 2026Physical AIPhysical AI needs a device contractWhen agents move from software tools to microscopes, robots and industrial instruments, interoperability must encode capabilities, safety boundaries and machine-verifiable state.
- Analysis26 Aug 2026AI EconomicsAgent FinOps starts before the invoiceAgent workloads turn AI spend into an operational systems problem: budgets, quotas, routing, concurrency and cost per accepted outcome must be designed before scale.
- Analysis26 Aug 2026Evaluation & ObservabilityOpenTelemetry may become the common contract for agent evaluationFramework-agnostic evaluation becomes possible when agent traces expose a common telemetry contract. That could separate quality assurance from orchestration vendors.
- Analysis25 Aug 2026Enterprise OperationsERP-native agents should start with exception-heavy workThe best first ERP agent is rarely a fully autonomous core process. Exception queues in procurement, logistics and finance offer bounded, measurable work with clear system truth.
- Analysis25 Aug 2026Industry IntelligenceVertical AI needs domain systems, not vertical promptsFinancial and legal AI illustrate the same rule: domain value comes from trusted data, permissions, skills, workflows and governance around the model, not vocabulary alone.
- Analysis21 Aug 2026Enterprise ArchitectureEnterprise software is becoming a capability layer for agentsSalesforce’s Headless 360 direction shows the shift from exposing raw APIs to packaging governed business capabilities for any authorised agent.
- Analysis21 Aug 2026Enterprise ArchitectureThe AI gateway is becoming the enterprise policy planeAs models, coding agents and MCP tools multiply, a common gateway can centralise routing, budgets, access, guardrails and telemetry without forcing one model vendor.
- Analysis20 Aug 2026Software EngineeringCoding agents are becoming managed enterprise infrastructureEnterprise coding agents now need identity, spend controls, audit, sandboxing and tool policy. The winning unit is no longer the IDE plugin but the governed development runtime.
- Analysis19 Aug 2026Governance & SecurityWeb-grounded agents need a source policyGrounding an agent on the public web creates a provenance problem. Domain, freshness and evidence policy should be part of runtime governance, not an afterthought.
- Analysis19 Aug 2026Governance & SecurityZero Data Retention becomes an architecture requirementOpenAI’s expansion of ZDR for eligible frontier-model API customers shows that retention, safety and observability must be designed as separate layers.
- Analysis18 Aug 2026Frontier RiskCyber-critical models change the release processWhen frontier capabilities cross critical cyber thresholds, model release becomes a security engineering process involving containment, monitoring and staged access—not just an API launch.
- Analysis15 Aug 2026AI TransformationFive levels of enterprise AI maturityA Lexdatum framework for distinguishing personal assistance from connected, automated, agentic and AI-native operations.
- Analysis14 Aug 2026Governance & RegulationText watermarking turns provenance into infrastructureModel-level text watermarking can add a provenance signal without replacing records, citations or human disclosure. Enterprises should treat it as one layer in a wider evidence stack.
- Analysis12 Aug 2026AI TransformationThe frontier gap: why some firms are moving from assistance to executionOpenAI’s enterprise data suggests that leading adopters are not merely using AI more—they are connecting it to longer, agentic units of work.
- Analysis7 Aug 2026Operating ModelTwo deployment models are emerging for enterprise agentsSalesforce platform data suggests a split between high-volume specialist agents and versatile multistep systems. Architecture should follow operational DNA.
- Analysis6 Aug 2026Governance & SecurityAgent guardrails must survive the whole workflowA tool call can be safe alone and dangerous in sequence. Stateful policies make agent governance depend on what happened before, not only on the current request.
- Analysis4 Aug 2026SecurityZero Trust for AI is a control-plane problemAgent security extends Zero Trust beyond users and devices to tools, memory, models, delegated identity and runtime actions that must be continuously governed.
- Analysis3 Aug 2026AI EconomicsModel prices fall. Architecture decides who captures the savingsAn 80% price cut for one GPT‑5.6 tier on Amazon Bedrock shows why model cost should be routed, measured and treated as a moving input.
- Analysis31 Jul 2026Governance & RegulationThe EU AI Act moves from roadmap to enforcementFrom 2 August 2026, enforcement and new transparency duties make AI inventory, disclosure and provenance operational—not theoretical.
- Analysis30 Jul 2026AI TransformationAI ROI starts when the work changesEnterprise AI returns become durable when teams redesign workflows, decision rights and operating models instead of merely adding assistants to unchanged work.
- Analysis29 Jul 2026Industry IntelligenceRetail enters the age of algorithmic decisionsAI agents are beginning to mediate discovery, comparison and execution. Retailers must become legible to both people and machines.
- Analysis28 Jul 2026InteroperabilityMCP grows up for productionThe July 2026 specification moves MCP towards stateless scale, explicit multi-round trips, background tasks and enterprise-managed identity.
- Analysis27 Jul 2026Enterprise ArchitectureOpen weights belong in the enterprise model portfolioOpen-weight models add deployment control, locality and continuity options. The right enterprise question is not open versus closed, but which workload needs which control surface.
- Analysis27 Jul 2026SecuritySecurity systems begin to perceive, reason and actMicrosoft’s Project Perception points to a multi-model, multi-agent cyber stack—and to the governance required when defence operates at machine speed.
- Analysis24 Jul 2026Frontier ModelsReliability floors matter more than benchmark peaksClaude Opus 5 illustrates the accelerating model frontier—but enterprise selection still depends on task economics, variance and verification.
- Analysis22 Jul 2026Customer OperationsFrom agents that answer to agents that resolveOpenAI Presence makes the production question explicit: value comes from reliable resolution, not conversational fluency alone.
- Analysis17 Jul 2026Agentic SystemsAgent platforms are becoming enterprise runtimesLong-running work, event triggers, human checkpoints and evaluations reveal the real shape of an enterprise agent platform.
- Analysis14 Jul 2026Enterprise ArchitectureThe model is not the platformClaude’s availability through Google Cloud illustrates a broader shift: frontier intelligence is increasingly consumed through enterprise control planes.
- Analysis8 Jul 2026Voice & MultimodalVoice is becoming an enterprise interfaceFull-duplex voice, tool use and provenance controls move conversational AI from call deflection towards real workflow execution.