C TWO

C TWO holds no robots of its own. It starts and stops processes, assigns capacity and reroutes work inside other vendors' automation platforms, and recovers them when they break. That includes restarting terminated processes, resetting virtual desktops after crashes and frozen sessions, and performing logins and logouts to switch users between processes, all authenticated writes into infrastructure and runtimes C TWO does not control.

One named third party system, Microsoft Power Automate, has a path in both directions through a connector C TWO publishes under its own publisher metadata. Outbound, C TWO "orchestrates and governs execution" of Power Automate flows the customer owns, while customers are responsible for their own Microsoft licenses. Inbound, a C TWO task exposed as a flow can be invoked from Power Platform. The connector needs an authenticated service account or personal access token and a reachable C TWO host.

Work is routed to a "bot, agent, API, or human", across every bot, AI agent, API and human workflow, and new technologies can be added without redesigning flows, with direct API calling a resource type of its own alongside robots. Apart from Microsoft, no third party automation platform is named. C TWO calls itself technology agnostic, and its proposition depends on supporting the major RPA vendors, but the current site says so in general terms, not by name. An Integration Hub is still advertised in older page metadata.

SourceMicrosoft Learn C TWO connector reference, ctwo.com/amp and rpasupervisor.com/our-solutions; learn.microsoft.com/en-us/connectors/ctworead 2026-09-12

C TWO calls Universal Orchestration "the operating system that manages work across every technology in a single, connected flow", and the platform "unifies every bot, AI agent, API, and human workflow into a single, orchestrated system". The unit of composition is the work item, an unusually fine grain.

Work is orchestrated at the item level and continuously routed to the best available resource on real time performance, cost and SLA pressure, through universal queues. The resource is chosen per item and can change as conditions move, instead of being fixed at design time. A work item can be handled by a robot, an AI agent, an API call or a person, chosen on the same criteria. New automations or technologies can be added without redesigning workflows.

Failure handling is part of the flow. When issues occur, the platform automatically recovers and reroutes work, "keeping processes moving without downtime, backlogs, or manual firefighting", with automated failover preventing lost work, so a flow reassigns its own failed steps to a different kind of resource. Governance rules sit over the flow, enforcing SLA priorities, security requirements and operational constraints across every process and every technology.

Sourcectwo.com/amp and ctwo.com/amp/enterprise-governanceread 2026-09-12

There is no knowledge grounding or retrieval, for a structural reason. C TWO is a control plane over work items. It decides which resource should handle a piece of work and when, on signals about performance, cost, SLA pressure and resource availability. It does not answer questions, and there is no user asking one.

The platform covers orchestration, governance and analytics, with no document store, corpus, index, ingestion path, embedding, citation or retrieval step, and nothing invites a customer to connect a knowledge base or upload content. What the platform holds is operational, not documentary, namely universal work queues, item level state, credentials, governance policies and historical execution data.

Predictive intelligence forecasts SLA risks and potential failures from execution history, which is a model reading telemetry, not retrieval over a knowledge base. Where knowledge grounding exists in a C TWO deployment, it belongs to someone else. The AI agents C TWO routes work to may well be grounded on the customer's documents, but they are the customer's agents built on other platforms. The product itself is moving quickly toward agent management.

Sourcectwo.com/amp and ctwo.com/amp/enterprise-governanceread 2026-09-12

People are a resource in the orchestration, not observers of it. Universal Orchestration assigns work "dynamically to the best bot, agent, API, or human" and unifies "every bot, AI agent, API, and human workflow" into one orchestrated system. A work item can be routed to a person as its handler, by policy, in the same queue as the machine resources, so the human step is part of the execution path. Approvals are named and logged.

Governance promises streamlined approvals in which humans are included only when needed, without creating bottlenecks, and every approval is logged, alongside every action captured and every decision visible. The guardrails are enforced and aimed at agent autonomy, since "As automation estates expand, and AI agents start making real decisions, governance becomes non-negotiable."

AI agents and bots operate within defined guardrails, unauthorized actions are blocked before they cause issues, and humans and digital workers operate "under the rules you set". The hold on an agent's action at run time is built into the platform and configured by the customer. C TWO shows no approval surface itself, with no view for the approver, no queue of pending approvals and no escalation path or timeout behavior, and the approval language sits in benefit bullets, not a described workflow.

Sourcectwo.com/amp/enterprise-governance and ctwo.com/ampread 2026-09-12

SOC 2 Type II is held, by C TWO's account, and controls are the product's own subject. Under the heading Proven Security and Compliance Standards, C TWO calls itself "SOC 2 Type II, DORA, and HIPAA certified", and a SOC Type 2 Compliant badge sits in the site footer. The wording is loose. DORA is an EU regulation and HIPAA a US statute, and neither issues a certification, so those two read as claims of compliance alignment.

SOC 2 Type II is a genuine third party attestation, backed independently by the footer badge. Customers get named controls. Unified credential management gives centralized control of all bot, agent and system credentials, which are secure, rotating and fully auditable, and policy driven workflows "use governance rules to enforce SLA priorities, security requirements, and operational constraints". Enforcement blocks instead of advising.

Unauthorized actions are blocked before they cause issues, and applications, environments and credentials are managed centrally, with policies applied consistently across every process. Credential rotation for machine identities is an uncommon control that matters here, since orphaned bot credentials are the standard failure of a scaled RPA estate. C TWO names no SSO, RBAC, SCIM or identity provider integration, has no trust center, and publishes no way to request the report.

Sourcectwo.com/amp/enterprise-governance and the ctwo.com site footerread 2026-09-12

End to end audit trails log every action, change, execution, escalation and exception across technologies automatically. The accountability commitment is four separate guarantees, that every action is captured, every decision is visible, every approval is logged and every change is accountable.

Every process and task has an end to end audit trail, and historical data stays accessible for compliance and reporting, for audits and operational handovers where "nothing is left undocumented". That answers what the automated worker did, not just why an answer was given. Live visibility sits alongside the log.

Real-Time Intelligence gives "end-to-end visibility into every workflow, across every technology and touchpoint", showing where work slows and surfacing bottlenecks, SLA risks and throughput constraints as they happen, with customizable dashboards for executives, CoEs or operations teams. Item level orchestration means every work piece is tracked across systems.

The trail is C TWO's own record, produced as the control plane about work it routed across technologies it does not own, and it exists because those underlying platforms each log only their own slice.

Sourcectwo.com/amp/enterprise-governance and ctwo.com/ampread 2026-09-12

What C TWO keeps is a record of work, and there is no agent memory. It processes "work items through universal queues" and tracks every work piece across systems at the item level, with "historical data accessible for compliance and reporting". That is a queue and its ledger, the record of what work existed and what happened to it. It persists because it is the system of record for the estate, the way a ticketing system or an ERP persists.

The same store, seen from the other side, is the audit trail, logging every action, change, execution, escalation and exception automatically. Nothing describes a named memory layer, a retained preference or correction the platform reapplies to later runs, context an agent carries across sessions, a stated scope or lifetime for anything held, or a way for customers to inspect, edit, export or purge such state apart from the work records.

The nearest thing to learning is routing on "real-time performance, cost, and SLA pressure", and predictive intelligence that "forecasts SLA risks and potential failures". Whatever the routing model retains is absorbed into an optimization model, not held as state a customer can retrieve. Where C TWO orchestrates a customer's AI agents, any conversational memory those agents hold is theirs and sits in the customer's estate.

Sourcectwo.com/amp, ctwo.com/amp/enterprise-governance and the Microsoft Learn C TWO connector referenceread 2026-09-12

Customers run C TWO in their own environment or in C TWO's cloud. In C TWO's own connector reference, host and base URL together route to "your own C TWO server, so that each customer connects to their own environment", either C TWO Cloud or on premises. Both shapes come with worked examples, a managed cloud host in the form connect24.ctwo.cloud and an on premises host with an optional port in the form ctwo.contoso.com:8443.

An on premises server must be reachable from Power Platform over HTTPS, with the firewall opened to the relevant outbound ranges. The choice is architectural, not a region toggle, and the API base path carries a division segment, so each tenant's environment is addressed separately in either shape. C TWO Cloud has no published region list or data residency commitment, and C TWO, headquartered in Bergen and selling to EU enterprises, makes no statement on EU hosting.

Sourcelearn.microsoft.com/en-us/connectors/ctworead 2026-09-12

Nothing prebuilt ships for a customer to adopt. Universal Orchestration, Enterprise Governance and Real-Time Intelligence are the three pillars of one Agentic Management Platform, described together as what the platform does, not as things a customer chooses among. Without Enterprise Governance, Universal Orchestration is not a whole product, just the same product with a hole in it. Nothing is separately installable, named as a unit or adoptable.

The product's own artifacts are configuration the customer writes. The customer enters SLA requirements per process, sets governance rules, and "creates as many underperformance profiles as required" so each process and resource group is handled to its own requirements. Dashboards are customizable per audience, again written by the customer, not selected. Connectors for common RPA platforms are how the platform reaches other systems, not packs.

No catalog, library, marketplace or contribution path appears on the site, and no page names prebuilt agents, playbooks, policy packs or industry templates. The solutions pages are split by role and industry, which is navigation, not content that ships. A C TWO Integration Hub offering "easy-to-use connectors" is still advertised in older page metadata on the governance URL.

Sourcectwo.com/amp, ctwo.com/amp/enterprise-governance and rpasupervisor.com/our-solutionsread 2026-09-12

Service levels replace scheduling, so work starts on conditions, not a clock. A customer enters business requirements per process, and the prioritization engine decides which resource works on what and when, starting and stopping processes to hold the service level. Work is "orchestrated at the item level and continuously routed to the best available resource based on real-time performance, cost, and SLA pressure". SLA pressure and queue state are themselves the trigger conditions.

Failures and exceptions drive a second path. When issues occur, the platform automatically recovers and reroutes work, with automated failover preventing lost work, and the events it handles include terminated processes it restarts, frozen sessions, crashed runtimes and unavailable virtual desktops it resets, and logins and logouts between processes. Escalations and exceptions are logged as events of their own. Outside systems can start work too.

A C TWO task configured as an exposed flow can be fired from Microsoft Power Automate, selected by Task ID from a dropdown in the calling system, so an event in a system C TWO does not own can start work inside it. Conversational channels such as chat, email, voice or ticketing are not part of the product and are not claimed. The customer is an automation CoE watching a queue, not an employee messaging an assistant.

Sourcectwo.com/amp, ctwo.com/amp/enterprise-governance and the Microsoft Learn C TWO connector referenceread 2026-09-12

Routing work to AI agents is not model choice. The platform assigns work "dynamically to the best bot, agent, API, or human" and compares RPA, APIs and AI agents on ROI, so a customer really is choosing which intelligent resource handles a work item. Those agents are the customer's own, running on the customer's own models in the customer's own estate. C TWO selects the worker but does not select or expose the model inside it, and it runs no inference of its own that a customer could point at a different provider.

C TWO does have intelligence of its own, an AI driven prioritization engine and predictive forecasting of SLA risk, but names no model, provider or routing choice for any of it, and there is no picker, admin setting or path to bring your own key. Nothing says C TWO routes across more than one model internally either. C TWO's own connector listing puts the product under Business Intelligence and AI and names no model provider.

Sourcectwo.com/amp and the Microsoft Learn C TWO connector referenceread 2026-09-12

Outside callers can reach C TWO through an API with a named authentication model. The marketing site at ctwo.com names no developer surface, and the knowledge base at support.ctwo.com is a Zendesk help center behind a sign in. A connection is configured against "the base path of your C TWO API including your division", given as /DEFAULT/CTWO.Server/api on the customer's own host.

Authentication is an API key sent as a bearer token in the Authorization header, with a service account token recommended for production and a personal access token for development and testing. Tokens are created in the product under Apps and connections, then Microsoft Power Automate, then Create service account, and the token is shown once. The inbound direction is explicit.

A C TWO task must be configured as an exposed flow before an outside caller can fire it, and if a trigger never fires, the advice is to confirm the task appears in the Task ID dropdown and is exposed. C TWO deliberately publishes its own tasks for a system it does not control to invoke. Detail is what is missing.

C TWO publishes no public endpoint reference, SDK or MCP server on any of its hosts, the API is described through one connector, not a developer portal, and the connector is marked Preview.

SourceMicrosoft Learn C TWO connector reference, publisher metadata naming C TWO, plus ctwo.com/amp; learn.microsoft.com/en-us/connectors/ctworead 2026-09-12

Operations are measured live, but nothing puts a change under test. Real-Time Intelligence measures execution continuously, covering where work slows and why across all systems, bottlenecks, SLA risks and throughput constraints, with real time adjustment of workloads, resources and priorities. Predictive intelligence "forecasts SLA risks and potential failures before they impact operations".

The closest thing to an evaluation is a real comparison, since the platform offers to "measure value across technologies" and compares RPA, APIs, AI agents and workflows for insights driven by ROI. A customer can read that comparison across resource classes and use it to decide whether an AI agent or a robot should own a process. But it measures resources as they run in production, not a change tried against a known set before release.

There is no test environment for orchestration policies, dry run of a routing rule, holdout, versioned policy with rollback or scored verdict on whether a change helped. A customer edits an SLA or a governance rule and it takes effect, with no gate between writing it and going live. Continuous rerouting on performance and cost optimizes resource allocation, not agent behavior, and the robots and agents being measured are the customer's own, which C TWO reports on but does not test. Self healing that detects and fixes bot failures is runtime recovery, not testing.

Sourcectwo.com/amp and ctwo.com/amp/enterprise-governanceread 2026-09-12

C TWO does not operate interfaces itself, even though the robots it manages do. RPA robots drive user interfaces for a living, clicking, typing and reading screens in applications with no API. C TWO dispatches, monitors and recovers those workers but never reads a screen. When a UI change breaks an automation, the robot breaks and C TWO is what notices, reports and reroutes the work. Its value is that it is not the brittle layer. Infrastructure remediation is the subtler case.

C TWO resets virtual desktops after terminations, frozen sessions, eternal loops and crashed runtimes, and performs logins and logouts to switch users between processes. That is administrative session management through the automation platform's own controls and infrastructure APIs, not driving a visual desktop. Restarting a machine is not using one. There is no hosted or local browser, no desktop session for the platform itself, no remote or local computer control for the customer, no recorder and no RPA authoring surface. C TWO deliberately does not build the robots, which is its market position.

Sourcectwo.com/amp and rpasupervisor.com/our-solutionsread 2026-09-12