Orchestration Is the Real Agent Gap
UiPath and Infobip

In the past two weeks, two very different companies delivered the same verdict. UiPath launched Maestro Flow on August 19; Infobip's AgentOS narrative, covered on August 27, points to the same conclusion: enterprises are not short of agents — they are short of the orchestration layer that turns agents from prototypes into governed business processes.

Two vendors, one verdict: the gap is orchestration

UiPath's Maestro Flow is a developer-first orchestration canvas: it lets coding agents — Claude Code, Cursor, GitHub Copilot, Codex — design, run, observe, and govern complete end-to-end business processes from their native development environment, as a single artifact from prototype to production. UiPath CPTO Raghu Malpani's line is widely quoted: "Enterprises don't have an agent problem; they have an orchestration problem. With coding agents, it's never been easier to build an agent. But running a real business process, spanning agents, robots, systems, and people — and being able to prove what happened at every step — needs an orchestration layer that ties these together in a real business environment."

Infobip's AgentOS is the same argument from a different direction. The Croatian communications company launched AgentOS on April 1 as a control layer for AI agents across marketing, sales, and support — connecting customer data, workflow automation, and channels like WhatsApp, SMS, RCS, email, and voice, then handing off to a human when the machine hits its limits. The commentary frames it as the next phase of enterprise AI being orchestration, not conversation: making the bot accountable for action, not just for dialogue.

Why coding agents build prototypes but cannot run processes

The context numbers show the size of the problem. BCG's 2025 research found 45 percent of APAC firms are experimenting with or deploying agentic AI; Adobe reports 88 percent of Asia-based respondents say fragmented data limits responsive, personalized customer experiences. Companies are not short of pilots — they are short of production systems. UiPath's diagnosis: coding agents make building trivially easy, but without an orchestration layer, process logic fragments, becomes hard to govern and costly to maintain — prototypes cannot ship, work stalls waiting on review, automations go untracked, and the cost of experimentation itself becomes a brake on progress.

Infobip's chief innovation officer Krešo Žmak makes the same layered point: the consequential layer sits around the model. Its Knowledge Agent answers from company documents via retrieval-augmented generation; the Agentic Chunker paired with a cross-encoder lifts vector-search recall from 70 to 95 percent while cutting document preparation from months to hours; Active Recall carries conversation history through a session; guardrails check agent output against business rules before it reaches a customer.

What the orchestration layer must provide: permissions, memory, audit, handoff

An effective orchestration layer needs four things, and skipping any of them breaks it. Permissions: role-based access control, data isolation, SSO, and API-level authentication — Infobip claims SOC 2 and ISO 27001 certifications, GDPR compliance, AES-256 encryption, web-interface actions logged for up to five years, and agent-layer tracing so a company can reconstruct why an AI responded as it did. Memory: cross-session context rather than starting from zero. Reliable retrieval: a governed knowledge base. Handoff: human approval and clean escalation paths — when the machine is uncertain or business impact is high, escalate rather than push through.

LAQO: from conversation to accountability

Infobip's flagship example is digital insurer LAQO. LAQO GPT runs on WhatsApp, grounded in policy documents, routing conversations by intent from the first message; an orchestrator was later added to coordinate a knowledge agent alongside two specialist agents for travelers and contractors. Per Infobip: the chatbot handled 30 percent of inbound requests, supported a 330 percent jump in sessions, and resolved 90 percent of queries within three to five messages; in the first half of 2026, LAQO GPT resolved about 40 percent of all customer interactions on its own, alongside a 4.7 customer satisfaction score and a 26 percent rise in completed travel insurance purchases. These figures come from the vendor's own client relationship — treat them as a case study, not an independent benchmark.

Three judgments for buyers

For digital transformation leads, these two weeks of product signal combine into one selection note: coding agents solve the build, orchestration solves the run. The former is already commoditized to near-free; the latter — permissions, memory, audit, handoff — is the real battlefield of the next round, and the hard gap for production-grade agents.

References

• UiPath press release: UiPath Launches UiPath Maestro Flow (2026-08-19) — https://www.uipath.com/newsroom/uipath-launches-maestro-flow
• Deeptech Times: The agentic AI battleground — who will control the enterprise orchestration layer? (2026-08-27) — https://deeptechtimes.com/2026/08/27/the-agentic-ai-battleground-who-will-control-the-enterprise-orchestration-layer/
• BCG 2025 research: 45% of APAC firms piloting or deploying agentic AI (via Deeptech Times); Adobe: 88% of Asia respondents cite fragmented data limiting personalization