About/Est. 2019

Two engineers who got tired of watching good teams pay for the same mistake twice.

Nimbus Works is Iris and Nathan. That's it — no juniors, no account managers, no "delivery team" you meet once and never speak to again. We take three clients at a time because that's how many we can look after properly.

01

The two of us

Both still do the work. Neither has been in a sales meeting this year that we didn't book ourselves.

Iris Bergland

Infrastructure engineer · Lisbon

Twelve years in infrastructure, most of it in networks and Linux before "platform engineering" was a job title. Iris spent five years on the platform team at a European payments company — 400 engineers, 60 AWS accounts, and a PCI scope that made every change a committee decision. Before that, operations at a managed hosting provider, which is where she learned what a real outage looks like when you're the one on the phone.

She does the network design, the Terraform that other people have to maintain six months later, and most of the cost work. She is the reason our reviews contain diagrams that are actually correct.

Focus: VPC and multi-account networking, IAM, infrastructure as code, Linux performance, cost architecture
Before: Platform engineering at a European payments company (Series E); ops at a managed hosting provider
Cloud: AWS (primary), GCP. Terraform and OpenTofu since 2016.
Languages: English, Norwegian, Portuguese

"The NAT gateway is not free: a field guide to egress"
"Terraform state is a database. Treat it like one."
"Why we stopped recommending Kubernetes to eight-person teams"

Nathan Okoye

Site reliability engineer · Toronto

Ten years in reliability, starting on the edge: three years at a CDN, where a bad config push could take out a few thousand customer domains in ninety seconds. Then reliability lead at a logistics marketplace, building on-call from scratch for a team that had never had a pager and was not enthusiastic about getting one.

Nathan writes the SLOs, runs the game days, and is usually the person who says the uncomfortable thing in the room — that the incident wasn't caused by the engineer who deployed, it was caused by a process that let a single deploy do that much damage.

Focus: SLOs and error budgets, incident command, observability, Kubernetes operations, on-call design
Before: SRE at a content delivery network; reliability lead at a logistics marketplace
Tools: OpenTelemetry, Prometheus, Grafana, PagerDuty, Kubernetes. Boring tools, on purpose.
Languages: English, French

"Your pager is a design document"
"The 4am test: does your rotation survive a bad week?"
"Postmortems we didn't write, and what happened next"

These pieces go out to clients and to anyone who asks — four or five a year, no newsletter, no tracking pixels. Email hello@nimbusworks.chenky.sbs and we'll send whichever ones you want.

Origin

We met in someone else's outage.

In 2016 we were both in a shared incident channel for a vendor we'd both bought from, at 2am European time, arguing about whether the problem was DNS. It was DNS. We kept talking afterwards, mostly to complain, and eventually to compare notes on the same three mistakes we kept seeing at different companies.

Nimbus Works started in 2019 with one client, a former colleague's Series A startup, and a rule we've kept: no engagement starts without a written scope and a date. We've been two people the whole time. Every time we considered hiring, we asked who would review the new person's work, and the answer was always us — so we didn't.

Founded2019. Registered in Portugal, invoicing in USD and EUR.
SizeTwo partners. No employees, no subcontractors.
CoverageEU mornings and afternoons, US East from 9am. We don't promise US West evenings.
Insurance€1M professional indemnity, certificate on request.
ClientsRoughly 30 engagements since 2019. Most are 8–60 people, SaaS or marketplace.

02

How we work

Five rules. They're not values on a slide; they're the reason we turn work down.

RULE 01

We tell you when not to spend

About one enquiry in four ends with us saying "don't do this project". It costs us revenue in the short term and it's the main reason clients come back.

RULE 02

Documents, not decks

You get markdown, PDF and the diagram sources. Nothing in a format you can't edit after we're gone, and no 60-slide deck standing in for an argument.

RULE 03

Read-only by default

We ask for a separate read-only role, not your admin keys. Write access happens in windows you schedule, and we log every change we make.

RULE 04

No bench, no bait

The person on the sales call is the person doing the work. If we ever can't staff something ourselves, we'll introduce you to someone and take nothing for it.