Agentic AI for insurance

Insurance runs on decisions. Now decisions also run on governed agents.

CriticalBridge is the agentic AI platform for P&C insurance: build end-to-end claims and underwriting workflows, embedded with digital workers, and govern every agent in your enterprise — including the ones you didn't build.

SOC 2 Type II Runs within your security perimeter Supports all major LLMs Scales to hundreds of agents
The problem

Task agents everywhere. Outcomes nowhere.

Insurers have bought AI by the tool: document ingestion here, a triage bot there, and vendor agents embedded in every SaaS platform. The result is standalone task agents multiplying across the value chain with no central governance, no end-to-end workflow, and no measurable ROI — while claims and underwriting teams get more tools and no better way to work.

That is not a technology gap. It is a platform gap.

The platform

One platform. From a single agent to the whole fleet.

Three products on one governed platform — so IT, the business, and the people doing the work optimize the same outcomes together.

BUILD

Agent Studio

Create, test, and deploy agents, skills, and end-to-end workflows across the insurance value chain — no-code, with connectors, policies, and human approvals built into the flow.

Explore Studio →
WORK

Agent Console

The industry workbench where people and agents handle claims and underwriting together — with digital workers doing the busywork and humans making the calls that matter.

Explore Console →
GOVERN

Agent Lighthouse

The control plane for every agent in your enterprise — native or third-party, any framework. Discover, enroll, enforce, and walk into any examination with the evidence already assembled.

Explore Lighthouse →
Digital workers

Digital workers, employed on your workflows.

Insurance-native digital workers run inside governed, end-to-end workflows: straight-through where the work allows it, guided-to-decision where a person should stay in charge — priced by the completed outcome, not by the seat.

Athena
Operations Agent

FNOL and submission intake, digitization, triage and assignment. Turns unstructured intake into structured, routed work.

Atlas
Underwriting Agent

Risk assessment through quote and issue. Full straight-through on personal lines and small commercial; guided-to-decision on complex risks.

Bella
Adjuster Agent

Low-complexity claims handled straight through; complex claims guided to a decision an adjuster approves. Salvage, subrogation, and SIU flags surfaced, not buried.

Open by design

Bring your own AI. Bring your SaaS agents. Keep your core.

CriticalBridge connects to the systems you already run and to whatever models you choose. Vendor agents from your SaaS platforms and homegrown agents on any framework join the same governed network.

GuidewireDuck CreekSalesforceServiceNowWorkdayClaudeGPTGeminiopen-source modelsLangChainCrewAIA2AMCP

Insurtech sells tools. SaaS embeds agents. CriticalBridge integrates, automates, and governs all of it.

Governance & trust

Examination-ready, by architecture.

Every decision an agent touches lands in an immutable, hash-chained audit trail with per-decision lineage — and every customer runs on their own isolated instance, so your blast radius is yours alone.

NAIC AI Bulletin
Evidence fields enforced on every regulated decision; examiner-ready export in one click.
EU AI Act
High-risk-ready control mapping on the same record fabric.
Per-customer isolation
Dedicated coordinator, database, and storage for every customer — as architecture, not an upsell.
Read how the trust architecture works →
Get started

Start with what you can see.

The Discovery Trial inventories every agent in your enterprise in 30 days — one read-only credential to start, metadata only, nothing deployed. Free, and you keep the Shadow Agent Report either way.

Home / Platform

An intelligent orchestration layer over the enterprise you already run.

CriticalBridge overlays your existing infrastructure — core systems, identity, data, models — with a governed agentic layer. Build, deploy, and govern AI agents and multi-step workflows within your security perimeter, without replatforming anything.

SOC 2 Type IISSO · RBAC · Audit Logs · Secrets MgmtSupports all major LLMs
Everything you need

Agents, workflows, and governance — unified.

Enterprise AI teams shouldn't need separate tools to build agents, design workflows, deploy integrations, and satisfy compliance. CriticalBridge is one environment for all four.

Agents

Enterprise agents

Create and deploy AI assistants, semi-autonomous agents, and autonomous agents — each with policies, controls, memory, and capabilities.

Workflows

No-code workflow builder

A drag-and-drop canvas for complex, deterministic workflows — validated at save time, auditable at run time.

Integrations

Hundreds of connectors

Salesforce, Slack, Google, AWS, Guidewire, databases, email, MCP servers, and a custom adapter path for everything else.

Enterprise-grade

Secure by design

SSO, RBAC, rules engines, immutable audit logs, secrets management, and field-level security — built in, not bolted on.

Observability

Analytics & monitoring

Real-time agent insights, performance metrics, SLA queues, and detailed execution logs across the whole fleet.

Governance

Risk-based controls

Risk-tiered classification with automated approval workflows, human-in-the-loop escalation, and examination-ready evidence.

How it works

From zero to governed automation in four steps.

01

Create enterprise agents

The agent wizard guides you through four phases — Define, Build, Govern, Validate — so identity, purpose, operational boundaries, policies, and approvals are set before anything ships. Start from the pre-built library or from scratch.

02

Add skills and workflows

An AI assistant helps generate reusable skills, inject organizational best practices, compose multi-step workflows, and wire integrations — so agents inherit your knowledge, not just your prompts.

03

Deploy in the right operating mode

Every agent runs in one of three modes: Assistant (advises and supports human decision-makers), Semi-Autonomous (executes the routine, escalates the exceptional), or Autonomous (operates independently within policy-bounded limits). The mode is a governance decision, not a model setting.

04

Coordinate the fleet with the Brain

The Brain Coordinator is the per-customer intelligence layer that balances workloads across agent pools, guides decisions with rules and AI, orchestrates multi-agent processes at enterprise scale, and enforces policies, compliance, and SLAs — one isolated coordinator per customer.

Under the hood

Even the model choice is governed.

A governed model router sits under every agent call: per-entity allowlists and denylists, data-residency constraints, quality tiers and cost modes, health-aware failover, and cascade escalation when a model underperforms — with an immutable, queryable audit answering “why this model” for every routed call. Route work across frontier APIs and open-source models under one policy. LLM cost control is a governance feature, not an afterthought.

  • Where people meet agents: the mesh routes conversations across Slack, Microsoft Teams, Google Chat, and the in-app panel — approvals arrive where the approver already works.
  • Open protocols: A2A is the backbone for agent-to-agent communication; MCP servers are first-class governed citizens.
  • Per-customer isolation: every customer gets a dedicated coordinator with its own database, cache, and storage. Provisioning tiers add a dedicated cloud project, customer-managed keys, and a private-IP data plane.
Home / Products / Agent Lighthouse
GOVERN

Every enterprise agent. One control plane.

See, govern, and orchestrate every AI agent in your enterprise — no matter who built it or what framework it runs on. Discovery that finds the agents you didn't know you had. Enforcement that checks every action before it runs. Evidence an examiner can consume.

Why now

Agents are multiplying. Control isn't.

Enterprises average a dozen agents today and are heading toward hundreds; analysts project 150,000+ agents per Fortune 500 company by 2028. Roughly half run in ungoverned silos — stood up by teams and vendors, invisible to any single inventory, with no common guardrails and no fast off-switch. Meanwhile the EU AI Act's high-risk obligations took effect in August 2026 and the NAIC AI Bulletin is adopted across half the U.S. states. Sprawl is the crisis; regulation is the deadline; the A2A protocol — the Linux Foundation standard backed by 150+ organizations — is what finally makes one control plane possible.

The model

A layer where we can't be accountable. A control plane where we can.

Lighthouse holds a line most vendors blur — and nothing moves between postures without a named human deciding it.

Tier 1 · Observed

Read-only discovery

Six surfaces — cloud & SaaS agent platforms, identity systems, network telemetry, code & CI, endpoints & MCP servers, and the A2A network itself — correlated into a live inventory with risk tiers and data-access maps. Nothing is touched, proxied, or interrupted.

Tier 2 · Sanctioned

Governed by record

Third-party agents you depend on but can't instrument get a named owner, continuous evidence refresh, and re-review the moment their posture changes. Accountable knowledge, honestly labeled — not fake control.

Tier 3 · Managed

Actively in the path

Agents you enroll get a conformance sidecar: policy evaluated at invocation time, execution rings with default-deny privilege ceilings, human approval on high-stakes actions, quarantine that revokes credentials, and per-decision lineage.

Instrumentation is what earns the right to enforce.

What every governed agent gains

Native governance, extended to every agent.

  • Fleet registration with risk tier and jurisdiction · heartbeat and liveness alerting
  • Performance metrics in dashboards and OpenTelemetry
  • Per-decision lineage with a signed authenticity envelope · immutable audit events
  • Policy evaluation before every action · human-in-the-loop approval gates
  • Just-in-time, scope-bound credentials · lifecycle control — pause, quarantine, promote, retire
  • A governed node in any cross-agent workflow
Differentiators

The claims we can defend, line by line.

A2A-native

Lighthouse governs agent-to-agent traffic on the open A2A standard, brokers cross-framework conversations, reads any agent's capability card — and detects when that card silently changes.

Discovery with published scopes

Six surfaces, read-only by construction, with the exact permission manifest for every connector published for your security team to review verbatim.

From finding to fixed

Discovery feeds one-click enrollment into the same governed fleet — not a findings report you still have to act on.

Fail-closed by default

No policy saved means no execution. New agents start in the sandbox ring, and anything not explicitly permitted is refused.

Evidence with integrity

Hash-chained audit records with server-recomputed verification and separation-of-duties break-glass — regulator-ready evidence, not just logging.

Plays well with what you run

Findings from installed security tooling ingest as first-class discovery sources. Bedrock stays Bedrock; Foundry stays Foundry; your IdP stays your system of record.

Insurance depth

Built where governance is mandatory.

Lighthouse operationalizes the NAIC Model AI Bulletin as enforced record fields, not a marketing label: five-factor harm classification sets logging depth and approval gates by use case; bias testing runs on schedules; consumer-notice justifications land on an append-only log; accountability names an owner, a backup, and an escalation path per domain; and one click exports an examiner-ready evidence pack reconstructing why any consumer received any outcome. The same field-surface pattern serves the EU AI Act and HIPAA.

See the full trust architecture →

See every agent you own in 30 days. Read-only. Reversible. Free.

One read-only credential to start; we widen the sweep in waves you approve. Never prompts, source code, secret values, or business data. First agents on screen about 30 minutes after the first connector. On day 30 you hold the Shadow Agent Report — yours to keep whether or not you buy.

Home / Products / Agent Studio
BUILD

Build agents and workflows with governance built in — not bolted on.

Agent Studio is where insurance workflows get built: a visual designer for end-to-end processes, a guided agent wizard with an AI authoring assistant, a skills builder, and a rules engine — with policy checks, approvals, and audit wired into every step at design time.

The wizard

Govern is a build phase, not an afterthought.

The agent wizard walks every new agent through four phases — so nothing reaches production without boundaries, policies, and validation.

01

Define

Establish agent identity, purpose, and operational boundaries. Set the business goal and desired results — or refine a template from the pre-built library.

02

Build

Configure capabilities, prompts, tools, and integrations. The AI authoring assistant reads, proposes, and validates as you go.

03

Govern

Set policies, approval gates, and compliance controls — risk tier, operating mode, and human-in-the-loop points are declared here, before deployment.

04

Validate

Test, assess risks, and finalize for deployment. Shadow mode runs proposed changes against production behavior; a propose → approve → promote pipeline gates every change.

Capabilities

Everything a builder needs, in one environment.

Workflow Designer

Compose deterministic, auditable workflows on a visual, no-code canvas — triggers, tasks, decisions, and actions with human handoffs where they belong. Workflows validate at save time and run on a deterministic engine, so what you designed is what executes.

Skills & best practices

Package expertise as reusable skills and inject organizational knowledge into agent decisions. Skills compose across agents, so what one team learns, every agent can use.

Any agent. Your rules.

Workflow steps can be executed by CriticalBridge agents, point-solution agents, and your SaaS platforms' agents alike — but only the agents you've enrolled, licensed, and cleared for that step are ever eligible, and the runtime refuses anything quarantined, retired, or outside its approved scope.

Connectors

Typed adapters for the systems agents actually touch: Guidewire ClaimCenter, databases, email, Slack, ticketing, HTTP/webhook, WebSocket, and MCP servers — with credentials vault-resolved and scoped per task.

Home / Products / Agent Console
WORK

Where adjusters, underwriters, and agents work the same queue.

Agent Console is the P&C workbench: claims from FNOL to resolution, underwriting from submission to issue, one environment over your core systems. Digital workers clear the busywork; your people carry the judgment; every handoff is governed.

The digital workers

Meet the team you don't have to hire.

Athena
Operations Agent

FNOL and submission intake, digitization, triage and assignment. Turns unstructured intake into structured, routed work.

Atlas
Underwriting Agent

Risk assessment, quote and issuance. Full straight-through on personal lines and small commercial; guided-to-decision on complex risks.

Bella
Adjuster Agent

Low-complexity claims handled straight through; complex claims guided to a decision an adjuster approves. Salvage, subrogation, and SIU flags surfaced, not buried.

Where people meet agents

Work doesn't happen in a dashboard.

The mesh routes agent conversations across Slack, Microsoft Teams, Google Chat, and the in-app panel — ask the concierge what the fleet can do, delegate a task to a named agent, and get the answer back in the thread, with the same policy gates and lineage as any other invocation. Human approvals arrive where the approver already works.

  • Work queue: SLA-sorted, with reassignment and override in operator hands
  • HITL approvals: high-stakes actions park for a person and resume on decision
  • Full context: every task carries its decision lineage, so reviewers see the why, not just the what
Economics

Priced by the outcome. Accretive on every unit.

Digital workers are priced per completed unit of work — an intake, a straight-through claim, a decisioned submission — not per agent and not per seat. Illustrative mid-size carrier modeling puts per-unit savings at roughly 75–80% across intake, straight-through claims, straight-through submissions, and guided complex work.

Home / Solutions / Claims

Claims, end to end — not task by task.

From FNOL to resolution in one governed flow: digital workers clear intake and triage, straight-through handling resolves the routine, and adjusters decide the complex with every fact in front of them.

Coverage

The whole claims value chain, on one platform.

FNOL & intake

Athena ingests notices from any channel, digitizes attachments, verifies policy, and opens a complete, structured claim — twice as fast as manual intake.

Triage & assignment

Every claim classified and routed by complexity and severity — full triage automation, with the Brain balancing workloads across adjuster and agent pools.

Investigation & evaluation

Agents gather statements, records, and estimates; rules and AI guide the evaluation; everything lands in the claim file with lineage for the adjuster review and decisions.

Resolution & administration

Low-complexity claims run straight through following rules; payments above threshold park for human approval.

Salvage, subrogation & SIU

Recovery and fraud signals surfaced automatically and routed to the right specialist with the evidence attached.

Adjuster workbench

One environment over Guidewire, Duck Creek — the queue, the file, the approvals, and the digital workers, together.

Governance built in

Every claims decision, examiner-ready.

Customer-facing decisions carry NAIC evidence fields, bias testing runs on schedule, and any payout above your threshold requires a person's approval — enforced by the platform, not by convention. When the market conduct exam comes, the evidence pack is one click.

How governance works →
Home / Solutions / Underwriting

Underwrite more business. Touch less of it.

From submission to issue in one governed flow: complete, structured submissions on arrival, straight-through quoting where the risk allows it, and guided decisions on the risks that deserve an underwriter.

Coverage

The underwriting value chain, on one platform.

Submission intake

Athena digitizes and ingests submissions, schedules, and loss runs — every risk arrives complete and structured, not as an inbox of PDFs.

Triage & assignment

Submissions scored and routed by appetite, complexity, and priority — the right level of risk touch, automatically.

Risk assessment

Atlas aggregates data across every submission and surfaces what matters, so underwriters make faster, better-supported decisions.

Quoting & issuance

Personal lines and small commercial run straight through: complex risks are guided to a decision the underwriter owns.

Account servicing

Establishing loss control service and processing mid-term policy changes handled by governed agents — with human approvals where rules require them.

Underwriting workbench

One environment over your policy admin core — the queue, the file, the insights, and the digital workers, together.

Governance built in

Underwriting decisions you can defend.

Risk and pricing decisions carry full lineage; bias testing runs against your book on schedule; and unfair-discrimination guardrails are enforced in the flow, not audited after the fact. The same fabric answers the DOI and the EU AI Act.

How governance works →
Home / Trust

Read-only, by construction. Isolated, by architecture.

One page for your security reviewer, your compliance officer, and your architect — what we can see, what we can never see, and how the evidence holds up.

Security architecture

What your security team will ask — answered up front.

Read-only, always

Discovery connectors never require write scope, and every connector ships the exact permission manifest — hand it to your security team verbatim.

Metadata only

Never prompts, source code, secret values, or packet payloads. We record that an agent exists, what it can reach, and who owns it.

Credentials never displayed

Write-only from the portal into the platform vault; refresh tokens never leave the coordination plane; agents receive just-in-time, scope-bound, short-TTL credentials.

It fails loudly

A broken connector shows unhealthy and raises a signal. It never quietly reports “no new agents” — silent breakage is the failure mode we designed against.

Per-customer isolation

Every customer runs a dedicated single-tenant coordinator with its own database, cache, and storage. Provisioning tiers add a dedicated cloud project, customer-managed keys with a kill switch, and a private-IP data plane.

SOC 2 Type II

Independently audited controls across the platform, with SSO, RBAC, field-level security, and immutable audit logs as standard.

Compliance depth

Frameworks as enforced fields, not marketing labels.

  • NAIC Model AI Bulletin — enforced evidence fields, bias schedules, consumer notice, accountability structure, shadow reports
  • EU AI Act — high-risk-ready control mapping on the same record fabric
  • NIST AI RMF 1.0 — alignment across the control set
  • Audit integrity — immutable, hash-chained records with server-recomputed verification and separation-of-duties break-glass access
Open standards

Trust you can verify from outside.

A2A is the backbone; MCP servers are first-class governed citizens; DID-based identity and signed trust manifests back the public AI Catalog — where a quarantined agent's public trust badge drops in real time. Published wire contracts are schema-pinned in CI, so federation peers never discover a broken contract in production.

Home / Company / About

Making agentic AI accountable for real business outcomes.

We're on a mission to make agentic AI accessible, scalable, and focused on delivering measurable business value — in the industries where governance is mandatory.

Our story

Born from a gap everyone could see.

Founded in 2025, CriticalBridge.AI began with a simple observation: despite near-universal AI adoption — 88% of organizations deploying AI in at least one function, enterprise GenAI spending surging past $37 billion — the overwhelming majority of enterprises are failing to translate investment into business value. Over 95% of GenAI pilots never deliver measurable P&L impact, and the agents that do ship run ungoverned.

Our founders had lived that gap at scale — advising the C-suite, leading national practices, and building enterprise technology. So we built the missing layer: a platform where business and domain experts, not just data scientists, can build, deploy, and govern AI agents and end-to-end workflows inside the enterprise security perimeter — starting with the industry where the stakes and the standards are highest, P&C insurance.

Values

What we hold ourselves to.

Customer-centric

We build features that solve real problems for real businesses.

Innovation

We push the boundaries of what's possible with automation.

Security first

Enterprise-grade security is built into everything we do.

Collaboration

Great products come from diverse teams working together.

Leadership

Leadership team

Frank Burkitt

CEO & Founder

Gail McGiffin

Chief Solutions Officer

Home / Company / Contact

Let's start a conversation.

Whether you're scoping a claims or underwriting workflow, want a working session on governance, or want to explore a partnership — we'd love to hear from you.

Message us

Send us a message

By submitting this form, you agree to our Privacy Policy.
How can we help?

Common reasons to reach out

Book a working session

A working session on your estate beats a canned demo — bring your claims or underwriting workflow and we'll map it live.

Start the Discovery Trial

See every agent you own in 30 days — read-only, reversible, free.

Partnerships

Core-system, insurtech, and security-tooling integrations — our discovery ingests your findings as first-class sources.

Careers & press

Join the team shaping governed agentic AI for insurance, or reach our media desk.

Home / Discovery Trial

See every agent you own in 30 days. Read-only. Reversible.

One read-only credential to start. We widen the sweep in waves, and you approve each one — so your estate comes into focus without a single change to your environment.

The 30 days

Three waves. You approve each one.

W0

Day one — zero new credentials

Already running an agent registry (Microsoft Agent 365, ServiceNow AI Control Tower, MuleSoft Agent Fabric)? We read that first — your existing inventory is on screen the same day, before you authorize anything new.

W1

Week 1 — one cloud account

A single read-only credential on your primary AI platform — Bedrock or Azure AI Foundry. One account, one owner to ask. First agents on screen about 30 minutes after the credential.

W2

Week 2 — identity & directory

Non-human identities holding LLM grants (Entra, Okta), correlated against Wave 1 — so ambiguous service principals resolve instead of piling up. This is where the unregistered and unowned agents appear.

W3

Weeks 3–4 — code & telemetry

Repos and CI for homegrown agents; agent traces in telemetry you already collect. Network and endpoint sweeps only if you want them. Every agent risk-ranked, with a data-access map.

Why it's easy to say yes

Engineered so security approves it in days, not months.

Read-only, by construction
View-only scopes only. We publish the exact IAM policy for every connector — check it line by line.
Metadata only
Never prompts, source code, secret values, or business data.
Nothing deployed
No sidecars, no gateways, no network taps. Discovery runs from our side of the API.
Revocable in one click
You issue the credentials; you revoke them. Your data package is deleted on request.

On day 30 you hold the Shadow Agent Report: your complete agent inventory — enrolled and unknown — risk-ranked with owners, data access, and an enrollment plan. Examiner-consumable, and yours to keep whether or not you buy.

Request access

Request your Discovery Trial

By submitting this form, you agree to our Privacy Policy.
Home / Company / Careers

Build the operating layer for enterprise agents.

We're a small, senior team turning agentic automation into infrastructure the Fortune 1000 will run on. If you want unusual scope, unusual pace, and unusually good teammates — read on.

How we work

Four principles you'll feel from day one.

Ship where it matters

We embed with customers, take the biggest problem on the table, and put working software in front of them fast.

Determinism over demos

Agents that impress a room and fail in production are worse than no agents. We measure, replay, and prove reliability.

Small teams, wide surface

Founding engineers own products, not tickets. If you want scope, you get it — and the responsibility that comes with it.

Honest defaults

Clear priorities, direct feedback, and calm postmortems. No performative urgency, no political overhead.

Open positions

Open positions

Pick a role to see the full description and apply. We read every application.

Don't see your role?

We hire against needs, not org charts. Tell us what you'd build.

Send an open application
Benefits

Compensation and support that respects your time and your craft.

Meaningful early equity

Real ownership in a category-defining company.

Competitive comp

Bands calibrated against top-tier enterprise SaaS.

NYC hybrid

Three days a week together in our New York office, two remote — collaboration when it counts, focus time when it doesn't.

Home-office stipend

Everything you need to build a durable, comfortable setup.

Health & wellness

Premium medical, dental, and vision for you and your family.

Learning budget

Books, courses, conferences — invest in your craft.

Generous PTO

Take the time you need. We track outcomes, not hours.

Onsite gatherings

Quarterly in-person weeks to align, ship, and connect.

Ready to build?

Pick a role or tell us what you'd build. Either way, we'll read it.

Home / Privacy Policy

Privacy Policy

Last updated: February 25, 2026
Home / Terms of Service

Terms of Service

Last updated: February 27, 2026