Atomicwork Adds Six Certified AI Coworkers to Its IT Marketplace

Atomicwork launched six pre-built, certified AI Coworkers in its AI Coworker Marketplace on November 6, 2026, targeting repetitive, high-volume IT service work with role-specific evaluations, verified integrations and controls for access, approvals, spending and audits.

The lineup pairs Atom, which handles employee IT requests in Slack, Teams and the portal, with Access Ops Manager for requests from intake through provisioning. Device Ops Engineer addresses endpoint issues through MDM and EDR, while Network Ops Engineer covers VPN, DNS, firewall and LAN/WAN problems. Cloud Ops Engineer handles cloud alerts, cost anomalies, and patch and backup verification, and SecOps Manager covers identity alerts, phishing, vulnerability triage and containment.

The six join a broader AI Coworker catalog listing more than 25 pre-built coworkers and over 50 connectors.

The certified release follows the June 9, 2026 introduction of Atomicwork’s AI Workforce, which established coworkers with defined scopes, skills, budgets and human managers. It shifts the offer from tools for constructing those roles to a pre-built lineup intended to reduce the integration and evaluation work customers must do themselves.

What Certification Tests

Certification has two parts. Atomicwork verifies that the tools, MCP connections and skills needed for a role are available and tested, then evaluates the coworker against enterprise scenarios built around simulated employees, preloaded company data and deliberate traps. The score examines whether the work reached the correct final state rather than judging only the first response. The evaluation framework also includes request metrics such as time to first response, mean time to resolution, first-contact resolution and SLA compliance.

Atomicwork says Atom is tested regularly against 638 scenarios spanning 12 support domains and 32 subcategories, with scoring for outcomes, evidence, execution, safety and communications.

The Cloud Ops Engineer’s product record shows how certification translates into a working role. The coworker includes AWS, Azure and Google Cloud Platform MCP connectors, cloud-specific diagnostics, triage and escalation skills, customer operating procedures and script libraries, an identity, a reporting manager and a budget. Atomicwork reports testing across more than 300 scenarios, with 25 graders per scenario and at least six domains. It is read-only by default, identifies what is not connected or cannot be checked, and routes actions beyond investigation through customer approval workflows. Its activity is logged as runs so reviewers can inspect what it checked, found and skipped.

Governance Continues After Deployment

Connected MCP tools are pinned and constrained by policy, while the platform enforces network, tool and code restrictions. Environmental changes require approval, and administrators can set monthly coworker budgets and inspect the cost of each ticket. Atomicwork says the certified coworkers can run on any harness and model and can be deployed inside customer infrastructure.

On-behalf-of access uses a token issued by the customer’s identity provider under its access policy, with every action audited. Atomicwork’s September 23, 2026 explanation of Cross App Access adds that read and write scopes are enforced on each call and that deactivating the Okta connection blocks the next token exchange.

Weekly drift checks compare production runs with expected behavior and recommend changes to goals, skills or tools. An administrator must apply a suggestion, run an evaluation and publish the new baseline only if the results hold. A September 15, 2026 engineering record describes judges that assess full trajectories for role-specific behavior, error handling and security signals such as tool misuse and privilege abuse. Proposed improvements do not allow a coworker to rewrite its own instructions freely or expand its permissions.

Availability and Marketplace Limits

The first certified coworkers are seeded as drafts in every Atomicwork account. Customers review a role, connect its tools, confirm its manager and budget, and then publish it. The AI Workforce can layer over an existing ServiceNow or Jira Service Management deployment without a migration. If a coworker cannot resolve an issue, it hands the case to its reporting manager or another human with the investigation and run record attached.

Atomicwork reports median turnaround times of 18 seconds for access automation when no approval is needed and 2.5 hours when approval is required. Those figures measure access automation in production, separate from the certification tests for the six roles.

Every coworker in the marketplace at launch is built and certified by Atomicwork. Outside contribution is limited to the skill layer through customer-supplied MCP, an MCP store and a skill registry. Publishing and certifying third-party coworkers against the same benches is a future plan. The release therefore pairs ready-made deployment with administrative supervision, while leaving a multi-vendor coworker marketplace outside the current product.