11+ years hardening enterprise environments across identity, network, endpoint, and detection. Now I build the runtime controls that let AI agents work with credentials, documents, payments, and production systems without standing access.
Results
Measured outcomes from e.l.f. Beauty, a $1B+ NYSE-listed public company, and from LiteWork, the AI automation agency I run today.
At e.l.f. Beauty · enterprise security · 2021–2025
endpoints
one dot ≈ 15 endpoints
Rolled out device compliance across e.l.f. Beauty on Okta FastPass + BeyondTrust EPM.
investigations led
one cell per investigation
Drove 125+ security investigations as Incident Response Lead, covering phishing, malware, and DLP escalations.
apps microsegmented
one cell ≈ 3 apps · grouped by risk tier
Replaced broad VPN access with Zscaler ZPA. 100% adoption in 8 weeks with zero productivity hit.
phishing rate
Dropped phishing click-rate over 12 months of global awareness work.
At LiteWork AI · my automation agency · 2025–present
production workflows
Built 60+ secure automation workflows now running in production, spanning service businesses, healthcare, real estate, legal, e-commerce, and beauty.
document workflows
Replaced manual review with validated document workflows and audit trails.
Story
Enterprise security taught me to make access explicit. LiteWork, my AI automation agency, turned that discipline toward client automation. Agents make the same problem harder: tools, secrets, permissions, and audit trails all have to hold up at runtime.
Story timeline
security → runtime → agents
four chapters · one discipline
Detection, DLP policy, SIEM triage, and live incident response.
Zero-touch deployment, CIS baselines, Okta SSO, and ITSM migration.
Led the IT and security integration of the $355M Naturium acquisition. Built device trust, ZTNA, privilege controls, and cross-signal detection across 1,100 endpoints.
Scoped tools, JIT credentials, OAuth rotation, and policy-gated agents.
Enterprise Programs
At e.l.f. Beauty I ran the programs that decide who and what gets access. Each one covered the whole company.
Identity, applications, privilege, egress, and telemetry share one operating model.
Passwordless sign-in on Okta FastPass with hardware-bound sessions and device posture checks. Stealing a password stopped being enough to get in.
Retired the corporate VPN. Each app is reachable only by the teams that need it, with least-privilege policy enforced at the broker.
Nobody keeps admin rights. Elevation is requested per task, risk-tiered through ServiceNow, and expires on its own.
Every connection leaving the company passes inspection before it goes out. Nothing leaves the edge unlogged.
A Python pipeline that fuses endpoint, vulnerability, and network telemetry into one investigation queue, mapped to MITRE ATT&CK.
Agent Security
At LiteWork, my AI automation agency, agents read documents, prepare financial data, and call business APIs for real clients. Software that acts on its own does not get the benefit of the doubt: no agent holds standing credentials, and every action passes seven gates before it touches a real system.
Workload identity without disk-resident credentials.
State-changing actions pass a deterministic gate first.
Agents only reach explicit, typed, allow-listed tools.
Credentials are brokered just-in-time with scoped access.
Public entry points stay separate from admin planes.
Per-task sandboxes limit lateral movement.
Agent actions emit traces, logs, alerts, and recovery points.
Case Studies
Two of those systems, end to end.
01
02
Agent Runtime Stack
Controlled execution needs more than a prompt: every request passes through policy, sandboxing, scoped credentials, and an audit path before it touches real systems.
Continuous operationPrimary control loop
Claude Code drives the primary inner loop against production systems. A hardened hook pipeline and scoped skill catalog provide pre- and post-execution policy enforcement before work reaches a real resource.
N-way fan-outParallel execution
Codex runs across isolated, ephemeral sandboxes, each with its own branch, diff surface, and human review gate. Tasks cannot see each other's state, and credentials are scoped per sandbox.
Runtime controls
The control model uses scoped tool modules, policy gates across the agent lifecycle, vault-backed credential delivery, and auditable action paths for workflows touching real systems.
Private runtimePlatform boundary
Cloudflare edge controls, private Tailscale access, Docker isolation, health checks, scoped runtime credentials, and rollback paths keep public entry points, admin paths, and automation runtimes separated by policy.
Credentials
Checkpoints across security fundamentals, cloud identity, and Apple device management.
Hiring, recruiting, or working on something I'd be a good fit for? Send a note, or connect on LinkedIn.
