$1.2M/yr
Recurring savings from AI-run back-office workflows at an energy services firm
From one agent to an AI workforce.
AI Factory is manufacturing discipline, productized. Built on the Brain. Aimed by the Twin. Gated by evals. Governed on every call.
What is SphereIQ AI Factory?
AI Factory is where enterprise AI agents are designed, built, evaluated, and operated: a visual and pro-code agent builder, multi-agent workflow orchestration, model management, an evaluation harness with pre-production gates, and a template library of agents proven in real deployments.
A factory has stations. Here are the six.
- Design
Agent builder
Visual builder with a pro-code escape hatch. Every agent declares its tools (MCP), knowledge scopes (Brain), memory (Engram), and policies (Governance).
- Orchestrate
Workflow builder
Multi-agent workflows with human-in-the-loop steps, triggered by events, schedules, or an email hitting an inbox.
- Test
Evaluation harness
Golden datasets, regression suites, LLM-as-judge scoring — and a Governance-enforced gate: no agent ships without passing its evals.
- Route
Model management
Registry, A/B testing, fallback chains, and cost-aware routing that feeds AI Economics Intelligence™. Anthropic-primary, never model-locked.
- Reuse
Agent Template Library
Start from agents proven in production, not blank canvases. Configure to your Brain and Twin; deploy in days.
- Operate
Run & observe
Full tracing on every run, utilization dashboards in the Executive Console, and per-agent cost and ROI attribution.
Anatomy of an agent
Drafts stay drafts until they pass.
Draft agent
Visual builder with a pro-code escape hatch.
Metadata-filtered knowledge.
Anthropic-primary, never model-locked.
No agent ships without passing its evals.
The publish gate
An agent cannot be published until it passes its golden-dataset evals at or above its accuracy threshold.
Draft agentgolden datasets · accuracy ≥ thresholdPublishedBlocked publishes are recorded on the audit ledger like everything else.
Draft agentgolden datasets · accuracy ≥ thresholdPublished
Blocked publishes are recorded on the audit ledger like everything else.
Evals measure citation hit-rate against ground-truth answers, with latency tracked per run.
Days, not months
Template-to-production time on an existing Brain + Twin foundation
0
Agents in production without a passing eval suite. The gate is enforced, not encouraged.
Twelve agents that already have jobs.
Each template comes from a real Sphere deployment — pre-wired to Brain knowledge scopes, Twin process maps, and Governance policies for its role.
AP Agent
invoice match & exception routing
Compliance Agent
policy & evidence checks
Technician Knowledge Assistant
field ops
Procurement Agent
intake to PO
Revenue Assurance Agent
leakage detection
HR Agent
policy Q&A & onboarding
CFO Agent
close support & variance
Content Pipeline
governed marketing ops
Manufacturing Agent
SOP & deviation support
Proven agent templates by department and vertical, cloned into draft agents your team then scopes and gates.
Get the Agent Template CatalogGoverned actions
When an agent's output triggers an action (email, Jira, Slack) —
01
Validation
02
Firewall
03
Policy
auto / requires approval / blocked
04
Approvals queue
05
Execution
Every step audited.
Asked before every Factory rollout.
Both, deliberately. Business teams use the visual builder; engineers drop to code with full SDK access. Either way, every agent declares the same tools, scopes, and policies — so governance doesn't depend on who built it.
Fallback chains route to alternate models or human queues, every run is traced end-to-end, and regression evals catch degradation before users do. Failure handling is designed in, because a workforce that fails silently is worse than none.
Factory works best on a Brain (knowledge) and Twin (targeting) foundation — that's why the chapters are ordered. But phased adoption is normal: many customers start with one template on one knowledge collection and expand from proof.
A workforce this capable needs exactly one thing above it.
Chapter 05: the reason your auditors, CISO, and CFO will say yes.