Ignite26/27
Back to all phases
Phase 04

January–February 2027

Operate

Operate is about staying alert while it runs. This is where the agents are assembled, connected to enterprise data via MCP and put into production. You learn where they fail, how you see it before the client does, and how to roll back without panic — while balancing cost, performance and quality under real constraints.

Learning objectives

  • Build an agent that plans, uses tools and handles state
  • Connect agents to enterprise systems via MCP with sound access control
  • Establish observability that gives insight into failures, latency and cost
  • Roll out changes safely with canary, feature flags and rollback
  • Secure the solution against prompt injection and govern agents in production

Core topics

  • 01Agent orchestration: planning, tool use, state management
  • 02MCP integrations against enterprise data and actions
  • 03GenAIOps: traces, observability, alerting
  • 04Rollout: canary, feature flags, rollback and kill switch
  • 05FinOps for AI: model routing, caching and distillation
  • 06Security and agent governance: injection defence, STRIDE, agent IAM

Technologies and frameworks

  • Langfuse dashboards · OpenTelemetry · Grafana
  • LaunchDarkly · GrowthBook (feature flags)
  • Vercel · Railway · Docker
  • STRIDE modelling · OWASP LLM Top 10

Sessions in this phase

  • Masterclass: 'where is it burning?' — observability
  • Dojo on production incidents from the cohort
  • Frontier Friday

Artefacts Ignitere deliver

  • 01Observability dashboard with SLOs and alerts
  • 02Runbook for incident response
  • 03Cost model for AI delivery with documented trade-offs
  • 04Threat model for an agentic solution

Courses in this phase

Shared coursesAll Ignitere

  • Agentic architecture and orchestration

    Build an agent that plans, uses tools and handles state.

  • GenAIOps and observability

    See what happens in production and catch the failures before the client does.

  • FinOps for AI services

    Send the right request to the right model. Cut cost without losing quality.

  • Security for AI agents: injection, STRIDE and MAESTRO

    Threat-model an agent solution as if an adversary is trying to break it.

Track courses · Ignite Builder

  • Azure deployment for Norwegian clients

    Cloud architecture that meets Norwegian data requirements and survives the security committee.

  • MCP integrations and custom server design

    Connect agents to enterprise data and actions with the Model Context Protocol.

  • AI deployment and infrastructure as code

    From notebook to production — build the infrastructure your AI solution actually lives in.

  • Agent deployment and harness in production

    From a working agent to an agent that survives production

Track courses · Ignite Tuner

  • Distillation and small models for cost and latency

    When training pays off, and when you should simply not.

  • Dataset craft and labeling

    Build golden datasets that genuinely teach the model — and the client — something.

  • Agent governance in production

    Identity, oversight and control once the agents go live

Track courses · Ignite Shaper

  • AI rollout and launch

    How to take an AI delivery from finished to actually in use

  • Operations and service agreements for AI solutions

    What the client is actually buying once the solution is delivered

Cross-cutting threads

  • AI economics and FinOps
  • EU AI Act practice
  • Consultant craft
Phase 03Build
Phase 05Adopt