Share

The Signal

Why Enterprises Need Scanners to Find Their Own Agents

By Rina Takahashi— January 28, 2026

Feature image for article: Why Enterprises Need Scanners to Find Their Own Agents

MuleSoft launched Agent Scanners in January 2026 to help enterprises answer what should be straightforward: what agents do we have running in production? When you need dedicated scanning tools just to see what you've deployed, the deployment wave has outpaced operational visibility.

While 79% of enterprises report adopting AI agents, only 23% are actively scaling them. Salesforce predicts hundreds of agents per employee in 2026, most sitting idle—"impressive but invisible." The gap between deployment capability and actual utilization isn't about adoption resistance or change management. It's about something more fundamental: whether deployed agents can be found by the people who need them.

The Signal

Why Enterprises Need Scanners to Find Their Own Agents

By Rina Takahashi— January 28, 2026

MuleSoft launched Agent Scanners in January 2026 to help enterprises answer what should be straightforward: what agents do we have running in production? When you need dedicated scanning tools just to see what you've deployed, the deployment wave has outpaced operational visibility.

While 79% of enterprises report adopting AI agents, only 23% are actively scaling them. Salesforce predicts hundreds of agents per employee in 2026, most sitting idle—"impressive but invisible." The gap between deployment capability and actual utilization isn't about adoption resistance or change management. It's about something more fundamental: whether deployed agents can be found by the people who need them.

Where This Goes

Who Manages the Agent When It Fails at 3am?

Organizations deploying agent systems keep hitting the same wall. Not technical limits. Not model capabilities. They can't figure out who owns the work when something breaks overnight.

The pattern shows up across enterprise deployments: agents handling web automation at scale fail in ways that look like workforce capacity problems, not software bugs. A session timing out isn't a ticket for engineering. It's a business process that stopped running. Someone needs to notice, diagnose, restart, and explain the gap to whoever was waiting on that output.

Most teams assume this is an infrastructure problem they'll solve with better reliability. Our read from operating thousands of concurrent web sessions: the infrastructure works fine. The org chart doesn't. Companies are deploying agents faster than they're defining who manages them when the work fails.

Where This Goes

Who Manages the Agent When It Fails at 3am?

Organizations deploying agent systems keep hitting the same wall. Not technical limits. Not model capabilities. They can't figure out who owns the work when something breaks overnight.

The pattern shows up across enterprise deployments: agents handling web automation at scale fail in ways that look like workforce capacity problems, not software bugs. A session timing out isn't a ticket for engineering. It's a business process that stopped running. Someone needs to notice, diagnose, restart, and explain the gap to whoever was waiting on that output.

Most teams assume this is an infrastructure problem they'll solve with better reliability. Our read from operating thousands of concurrent web sessions: the infrastructure works fine. The org chart doesn't. Companies are deploying agents faster than they're defining who manages them when the work fails.

TAKE NOTE

Accountability gaps: Agents with defined ownership and clear escalation paths see materially higher adoption than shared resources managed by engineering teams.

Management models: Domain experts increasingly manage agent configurations directly, adjusting instructions and scaling successful patterns without developer involvement.

Failure classification: When agents run business processes at workforce scale, outages require operational response protocols, not just incident management systems.

Interface requirements: Non-technical managers need platforms that expose agent behavior and performance without requiring API knowledge or technical debugging skills.

Production reality: The gap between pilot and scale isn't technical capability but organizational readiness to own agent work as business operations.

From the Labs

What Production Agent Systems Actually Look Like

Survey of 306 practitioners building systems that serve millions daily reveals production agents favor simplicity over sophistication. Most execute under 10 steps, run on off-the-shelf models, and depend on human evaluation. The primary challenge remains reliability, not architectural complexity.

Where's the disconnect?

Research prototypes chase complexity while production systems prove simple, controllable approaches already deliver measurable impact.

What enables scale?

Human oversight and straightforward architectures, not elaborate coordination schemes, power agents across 26 industries today.

When Multi-Agent Coordination Helps Versus Hurts Performance

Controlled evaluation across 180 configurations quantifies three effects that matter: tool budgets create coordination tradeoffs, capability gains plateau above 45% baseline, and topology choices amplify errors by factors up to 17x. Sequential reasoning degrades 39-70% under multi-agent approaches.

Can we predict optimal strategies?

A quantifiable model lets architects choose coordination patterns based on measurable task characteristics, not intuition.

Which tasks benefit?

Parallelizable work improves 80.8% with centralized coordination, but sequential reasoning consistently degrades across all multi-agent variants.

Systematic Framework for Diagnosing Agent System Failures

First systematic mapping connects agent architectural components to observable behaviors and evaluation metrics. The architecture-driven approach addresses ad hoc evaluation practices by enabling targeted debugging. Practitioners gain clarity about which specific component fails rather than just detecting that something broke.

Why traditional testing falls short?

Test-driven development assumes determinism, but probabilistic components demand fundamentally different diagnostic approaches for production systems.

What changes operationally?

Linking behavioral signals to specific components transforms failure detection into actual diagnosis of root causes.

Engineering Lifecycle for Production-Grade Agent Workflows

End-to-end guide introduces structured lifecycle spanning workflow decomposition, multi-agent patterns, and deterministic orchestration. Nine core practices emphasize tool-first design, single-responsibility agents, externalized prompt management, and containerized deployment. Addresses the prototype-to-production gap through concrete architectural guidance.

What's missing from prototypes?

Governance, observability, and operational reliability that enterprises require before deploying agents at scale.

How does architecture help?

Single-responsibility agents and deterministic orchestration provide concrete patterns for tackling operational challenges systematically.

From the Labs

What Production Agent Systems Actually Look Like

Survey of 306 practitioners building systems that serve millions daily reveals production agents favor simplicity over sophistication. Most execute under 10 steps, run on off-the-shelf models, and depend on human evaluation. The primary challenge remains reliability, not architectural complexity.

Where's the disconnect?

Research prototypes chase complexity while production systems prove simple, controllable approaches already deliver measurable impact.

What enables scale?

Human oversight and straightforward architectures, not elaborate coordination schemes, power agents across 26 industries today.

From the Labs

When Multi-Agent Coordination Helps Versus Hurts Performance

Controlled evaluation across 180 configurations quantifies three effects that matter: tool budgets create coordination tradeoffs, capability gains plateau above 45% baseline, and topology choices amplify errors by factors up to 17x. Sequential reasoning degrades 39-70% under multi-agent approaches.

Can we predict optimal strategies?

A quantifiable model lets architects choose coordination patterns based on measurable task characteristics, not intuition.

Which tasks benefit?

Parallelizable work improves 80.8% with centralized coordination, but sequential reasoning consistently degrades across all multi-agent variants.

From the Labs

Systematic Framework for Diagnosing Agent System Failures

First systematic mapping connects agent architectural components to observable behaviors and evaluation metrics. The architecture-driven approach addresses ad hoc evaluation practices by enabling targeted debugging. Practitioners gain clarity about which specific component fails rather than just detecting that something broke.

Why traditional testing falls short?

Test-driven development assumes determinism, but probabilistic components demand fundamentally different diagnostic approaches for production systems.

What changes operationally?

Linking behavioral signals to specific components transforms failure detection into actual diagnosis of root causes.

From the Labs

Engineering Lifecycle for Production-Grade Agent Workflows

End-to-end guide introduces structured lifecycle spanning workflow decomposition, multi-agent patterns, and deterministic orchestration. Nine core practices emphasize tool-first design, single-responsibility agents, externalized prompt management, and containerized deployment. Addresses the prototype-to-production gap through concrete architectural guidance.

What's missing from prototypes?

Governance, observability, and operational reliability that enterprises require before deploying agents at scale.

How does architecture help?

Single-responsibility agents and deterministic orchestration provide concrete patterns for tackling operational challenges systematically.

Quiet Tech That Compounds

Model benchmarks move. Framework launches generate buzz. Meanwhile, something quieter compounds.

The infrastructure layer is maturing in ways that don't make headlines. Standardized protocols. Execution primitives. Deployment patterns that work when thousands of agents run simultaneously. When Create.xyz spins up 20,000 databases in 36 hours, you need infrastructure built for that reality. When your observability has to trace multi-step reasoning across frameworks, you need conventions that actually work.

This is foundation work. What makes production possible.

Quiet Tech That Compounds

Model benchmarks move. Framework launches generate buzz. Meanwhile, something quieter compounds.

The infrastructure layer is maturing in ways that don't make headlines. Standardized protocols. Execution primitives. Deployment patterns that work when thousands of agents run simultaneously. When Create.xyz spins up 20,000 databases in 36 hours, you need infrastructure built for that reality. When your observability has to trace multi-step reasoning across frameworks, you need conventions that actually work.

This is foundation work. What makes production possible.

Observability Standards

OpenTelemetry Conventions Define Agent Tracing

OpenTelemetry finalized semantic conventions for AI agents. LangGraph, AutoGen, and CrewAI traces now work with existing APM tools. When 89% of organizations already run observability infrastructure, standardized agent metrics slot right in. No new toolchain required.

Integration Protocol

MCP Becomes Universal Tool Connection Standard

Anthropic's Model Context Protocol got adopted by OpenAI, Google, and Microsoft. Stripe, Neo4j, and Cloudflare already ship MCP servers. One interface for tools and data sources across providers. Integration silos are dissolving faster than anyone expected.

Data Infrastructure

Sub-Second Database Provisioning for Agent Workloads

Agents create databases at 4x human developer rates. Create.xyz spawned 20,000 Neon databases in 36 hours. Legacy systems built for human timescales can't handle it. New infrastructure provisions instantly, scales automatically, isolates environments by default.

Execution Isolation

Kubernetes Agent Sandbox Provides Kernel-Level Isolation