The challenge behind the engagement.
AI Agent & Non-Human Identity is for environments where service accounts, machine credentials, and agents outnumber people. We inventory every non-human identity across the platforms you run, from directory and cloud identities to automation, delivery-pipeline, and application-to-application credentials. Each entry gets an owner, purpose, credential type, last use, and permissions, then a risk score against the ways non-human identities are commonly abused. We design the move from stored secrets toward certificates and platform-issued identities that need no secret at all. Agent identities follow an emerging pattern: one identity per agent, clear isolation boundaries, and a named sponsor. You receive the inventory, a risk register, a migration plan, and a lifecycle standard. Rotations are sequenced so production jobs keep running.
For platform, operations, and security engineering leads with hundreds of service accounts and access keys. Also for teams putting agents into production, whatever they are built on, who must decide whether each agent gets its own identity or acts as the signed-in user.
What we do.
What you can use.
Build the non-human identity inventory
We pull identity reads from each platform, workload sign-in and cloud audit logs, and repository scans for hard-coded secrets, then interview owners. The result lists every workload, service account, automation, and agent with owner, credential type, credential age, permission level, and last use. Entries without an owner become decommission candidates.
Design credentials and agent identities
We move stored secrets toward certificates, platform-issued identities, and pipeline-to-cloud trust that needs no stored secret. Where a service mesh exists, workloads get short-lived issued identities. Agents get one identity per logical agent, one isolation boundary per pattern, a named sponsor, and access templates for autonomous and on-behalf-of use.
Migrate, rotate, and prove it
The migration plan sequences each credential change with the job that depends on it, a test window, and rollback. After your team executes, we re-read sign-in logs and credential lists to confirm old secrets are unused and removed, and hand over the lifecycle standard and offboarding runbook.
Who it’s for.
When you need it.
- Platform and DevOps teams whose service accounts and machine credentials outnumber human users
- Organizations putting AI agents into production that can act without a human present
- Environments with long-lived cloud access keys and secrets scattered across repositories and pipelines
- Teams with no owner, inventory, or lifecycle for their non-human identities
- An agent or automation rollout is scheduled that needs its own identity
- A leaked secret, scanner alert, or audit surfaced unmanaged machine credentials
- A departing engineer's personal tokens turn out to run production jobs
- A key-expiry or rotation deadline threatens to break unattended automation jobs
What the scope can include.
- 01
Inventory of workloads, service accounts, agents, access keys, and pipeline identities across platforms
- 02
Owner, purpose, credential age, last use, and permissions recorded for each non-human identity
- 03
Risk scoring against the common failures of non-human identities, including long-lived secrets and human reuse
- 04
Credential migration from stored secrets to certificates or platform-issued identities
- 05
Agent identity design: isolation boundaries, sponsors, autonomous versus on-behalf-of permissions
- 06
Token handling for agents and their connectors: bound audiences, short lifetimes, no pass-through
What it typically costs.
One rate: $150/hour.
Every engagement is priced by the hours it takes at one flat rate, with scoping, the work, and the final deliverables included. Find the size closest to yours.
One cloud platform, a few hundred machine accounts and keys, a handful of AI agents
About 40–72 hoursCloud, directory, and build systems, thousands of machine accounts
About 100–160 hoursMany platforms and teams, AI agents acting on production systems
About 180–300 hours- Number of platforms where machine accounts and keys live
- How many service accounts and agents lack a known owner
- Whether stored secrets must be replaced with short-lived credentials
- Number of AI agents with access to real business systems
2–4 weeks for a single platform; 6–10 weeks across many platforms
Get a fixed quote for your scopeRanges are planning estimates at $150/hour, not a quote. Your price is confirmed in writing after a scoping call, before any work begins.
What you take forward.
- Non-human identity inventory with owner, credential type, age, permission level, and last use for every entry
- Risk register mapped to the common non-human identity failures with a fix per finding
- Agent identity design with isolation boundaries, per-agent identities, permission baselines, and access templates
- Lifecycle standard, offboarding runbook, secrets standard, and evidence of rotations and removals after remediation
Final coverage, deliverables, timing, and any retesting or implementation work are confirmed before the engagement begins.
Before we get started.
Should our agents get their own identity or run as the signed-in user?
Usually both, split by pattern. An agent that runs on its own schedule gets its own identity with its own granted permissions, so every action is attributed to it. An agent working inside a user session acts on that person's behalf and can never exceed what they can do. An agent gets a full user account only when it needs a mailbox or chat presence. Each logical agent gets one identity, grouped by isolation boundary, because a compromised boundary affects every agent beneath it. Governing agents with access policies usually needs specific licensing.
Can you rotate credentials without breaking production jobs?
Rotation is sequenced, not bulk. For each credential we identify the jobs that use it from sign-in and audit logs, then add the new credential alongside the old one. The job cuts over in a test window, and the old credential is removed only after it shows no use. Where a workload supports it, we move to a platform-issued identity so there is no stored secret to rotate again. Stronger sign-in requirements typically exempt true machine identities but not personal accounts used as service accounts, so those move first.
- NIST SP 800-63-4: Digital Identity Guidelines
- NIST SP 800-207: Zero Trust Architecture
- Zero Trust Maturity Model | CISA
- OWASP Non-Human Identities Top 10
- Securing privileged access Enterprise access model - Privileged access | Microsoft Learn
- What is Microsoft Entra Agent ID? - Microsoft Entra Agent ID | Microsoft Learn
- RFC 9700: Best Current Practice for OAuth 2.0 Security