Notes · Seven one-page notes

What we learned
on real builds.

Each note comes from a build we did, with the decisions and the numbers that mattered. Nothing here names a client who has not agreed to be named.

87%Alarm noise suppressed
across 250 situations
42sMedian root-cause analysis
in a Tier-1 NOC, from 2 to 4 h
9AKS enablement sessions
for a municipality's team
8×H100Sized for a roughly
400B-parameter model
Contents

Seven notes, three topics.

Before you commission a review

Which question needs an answer?

What does a landing-zone assessment cover?

It examines the agreed estate's identity, network boundaries, policies and operating evidence. You receive findings tied to resources, priorities and owners. Start here when the problem crosses teams or subscriptions. Confirm the estate boundary and access before the review starts.

Explore the landing-zone assessment →

When is a Kubernetes readiness review useful?

Use it before a production launch or handover when cluster controls, workload defaults and operating procedures need scrutiny. It helps identify readiness gaps and next actions. Agree which clusters and supporting evidence are included; a configuration check alone cannot establish recovery behaviour.

Explore the Kubernetes readiness review →
01 of 07 · Network operations

What supported the move from NOC MVP to production

Three things, from pilot to production.

87%Alarm-noise suppression
42sMedian time to root cause
84%Notes accepted as written
250Historical situations measured

At a Tier-1 Nordic operator, a multi-agent pipeline now turns fault situations on the 4G/5G Packet Core into ranked root causes and a recommended action. The MVP took six months for six use cases, and the operator funded production on the measured results.

01

Flows the NOC can audit

A deterministic four-step flow (retrieve, annotate, recommend, format) instead of an open agent loop. Every output is reproducible, so engineers could check the agent's work. They accepted 84% of its notes as written.

02

Inference inside the tenancy

Public models were ruled out on network-data confidentiality. An open-weight model runs behind a gateway on the operator's own Kubernetes, with the embeddings on persistent storage. Tracing runs end to end, and the operator's identity provider sits at the gateway.

03

A KPI baseline agreed with operations first

Alarm-noise suppression and time to root cause were measured against 250 historical situations before anyone argued about the design. The numbers that came out, 87% suppression and a median of 42 seconds, are the numbers that funded production.

02 of 07 · Network operations

From observability to agentic assurance

Three decisions that scope the first use case.

6 in 6Use cases in months
~1,000Fault situations a quarter
2 to 4 hManual baseline per situation
Week 1KPI baseline signed off

At the Tier-1 Nordic operator, fault situations on the 4G/5G Packet Core reach the NOC through BMC Helix. That fault data is the raw material the situation enrichment agents work on, and three decisions decide whether an agent earns a place in the NOC.

01

Which situations the agent may enrich

We started on the Packet Core only, with three use cases on each of two network element types, and widened the scope only after the acceptance rate held.

02

The agent never acts on the network

It recommends, and the NOC decides. An open loop keeps accountability for the network with the engineers who already carry it.

03

Who signs off the KPI baseline

Operations, in week one, on historical situations. If the people who will measure the agent did not agree the measure, the pilot will not become production. The numbers it produced: note 01.

03 of 07 · Network operations

Agents and BMC Helix over A2A

Integrating agents with the operations backbone.

The design covers
Submit and pollAgent cardAudit recordReplay
The transition
RESTA2A

At the Tier-1 operator, BMC Helix is the operations backbone. Fault situations arrive from it, and the agents hand ranked root causes back to it. We designed that integration against HelixGPT's Best Action Plan using the A2A protocol.

01

Every exchange leaves a record

One audit record per exchange, and replay when something fails mid-flight. Every recommendation that lands in a ticket can be traced back to its inputs.

02

Why REST first, then A2A

The agents started on a REST integration and moved to A2A in step with the platform release that supported it. The operator never ran an unsupported path.

03

What we would do again

Keep the agent tools on MCP and the agent-to-agent traffic on A2A through a gateway with the operator's identity provider. The two protocols do different jobs, and mixing them costs you the audit trail.

04 of 07 · Cloud platforms

What makes a Kubernetes platform stick

Three things that decide it, across multi-team platforms.

3Clouds, one set of defaults
3Kubernetes services standardised (AKS, EKS, GKE)

We have standardised managed Kubernetes for an operator across AWS, Azure and GCP, and assessed and enabled a Dutch municipality's AKS landing zone this year. The pattern of what sticks is the same in both.

01

Ingress, egress and access tiers before the first workload

Decided once, centrally, so distributed teams start with resilience and security in place rather than retrofitting them under delivery pressure.

02

Guardrails as self-service modules, not gates

A control that needs a request gets bypassed. A Terraform or Bicep module a team can consume on its own gets adopted, and compliance runs in the pipeline instead of in a review queue.

03

A plan for the people who inherit the platform

A platform the in-house team cannot operate is not finished. The full note describes the self-service modules we used.

05 of 07 · Cloud platforms

Landing zones for public sector and healthcare

Four things that carried over between two platforms.

9Enablement sessions
2Platforms through one MSP partner
Day 1Monitoring baseline

Through an MSP partner we assessed a Dutch municipality's AKS landing zone and built a secured Azure Virtual Desktop platform for a national healthcare screening organisation. Four things carried over between them.

01

Assess with evidence from the estate

Resource Graph and Traffic Analytics, not interviews alone, so every finding points at a resource someone can fix.

02

Identity before desktops

The screening organisation's Entra ID and Kerberos gaps were closed into a secured landing zone before the desktop platform went in. A desktop platform is only as safe as the identity it trusts.

03

A monitoring baseline on day one

Azure Monitor alerting shipped with the platform, so nobody discovered the missing alerts in an incident.

04

Enablement as a programme

Nine sessions for the municipality's engineers, because the platform is not finished until the in-house team can run it.

06 of 07 · Agentic applications

Keeping agents auditable

Guardrails for an agent runtime.

Execution order
IntentTool chainConfirmationAction

These are the guardrails we run in our own real-time voice agent runtime, and the ones we design into agents for clients. None of them depends on the model behaving.

01

A deterministic execution model

Intent, then tool chain, then confirmation, then action. Flow drift from the model is replaced by a stateful runtime that decides what runs.

02

One tool abstraction

The model can only invoke the tools it has been given through a single, governed interface. It cannot reach a system it was not handed.

03

Tracing end to end

OpenTelemetry on every step, with latency objectives on time to first byte and between chunks, and drift detection in production.

04

Compliance as architecture

GDPR session boundaries and data minimisation, and disclosure requirements translated into conversation-flow decisions for review with the responsible legal and privacy teams.

07 of 07 · Agentic applications

Sizing an air-gapped GenAI platform

And the residency question.

2Planes, CPU and GPU
8×H100For a roughly 400B-parameter MoE model
35BCosted alternative
2Options, both costed

For a Nokia account team in Japan we advised on an on-premises, air-gapped GenAI platform for network-function documentation. Nothing could leave the customer's site, and the hardware had to be defended before it was ordered.

01

Two planes

A CPU plane for everything that is not inference, and a GPU plane reserved for it. CPU capacity is cheaper to add, and the GPU plane should never be shared with retrieval or orchestration.

02

Two sized options

A roughly 400B-parameter open-weight mixture-of-experts model on 8×H100, and a 35B-parameter alternative, both costed, so the customer could weigh answer quality against hardware.

03

Embeddings evaluated, not assumed

Retrieval quality over the documentation decides whether engineers use the platform at all.

04

Residency and sovereignty kept separate

An air-gapped platform satisfies both, but each customer requirement was traced to one or the other, so the design could be defended line by line.

Ask for the full note.

The full one-page notes are free on request. Tell us which one, or bring the problem it is closest to, and we take it from there in a 45-minute working session. Bring the problem and your questions; we will clarify the next useful step.

Request a working session → See the Work